more env fixes

README.md

@@ -54,11 +54,11 @@ All implementations read `model.cfg` (or `.bantam.cfg`, which takes priority if
    ./mb              # MicroBantam (Perl 5)
    ```
 
-   In interactive mode, prompts can span multiple lines: press **Ctrl+J** to insert a real line break (the cursor moves to the next line), then **Enter** to submit the whole multi-line prompt. The Go port ships its own raw-mode line editor (arrow keys move the cursor, Up/Down browse history, Backspace edits, Ctrl+C clears line / interrupts in-flight run, Ctrl+D exits), working everywhere without third-party dependencies.
+   In interactive mode, prompts can span multiple lines: press **Ctrl+J** to insert a real line break (the cursor moves to the next line), then **Enter** to submit the whole multi-line prompt. The Go port ships its own raw-mode line editor, working everywhere without third-party dependencies. It supports arrow keys and Up/Down for history, Backspace and Ctrl+D (non-empty line) to delete, and Emacs-style editing combos: **Ctrl+A** (start of line), **Ctrl+E** (end of line), **Ctrl+B** / **Ctrl+F** (move by character), **Ctrl+W** (delete previous word), **Ctrl+K** (kill to end of line), **Ctrl+U** (kill to start of line), **Ctrl+Left** / **Ctrl+Right** (move by word), and **Home** / **End** keys. **Ctrl+C** clears the line / interrupts an in-flight run, and **Ctrl+D** on an empty line exits.
 
    After every interaction, Bantam displays token usage (prompt tokens, cached/uncached breakdown when supported by the provider, completion tokens, and context window utilization):
    ```text
-   [tokens: 1420 prompt (1000 cached, 420 uncached) + 85 completion | context: 1420/200000 (0.7%)]
+   [openrouter/free: 1420 prompt (1000 cached, 420 uncached) + 85 completion | context: 1420/200000 (0.7%)]
    ```
 
    Sessions are saved under `~/.bantam/sessions/` and can be managed with these commands:
@@ -68,6 +68,9 @@ All implementations read `model.cfg` (or `.bantam.cfg`, which takes priority if
    - `/load <id>` — load a saved session (exact id or unique prefix) and continue from there
    - `/compact` — compact context down to the system message and a concise summary using the LLM; the compaction prompt is appended to the conversation to derive the summary, then the conversation is reset to `[system, summary-user-message]` (a fresh prefix, so downstream prompt-cache hits depend on the provider and are not guaranteed)
    - `/cfg <param> [val]` — inspect or update a configuration parameter live (writes to `.bantam.cfg`)
+   - `/model [val]` — alias for `/cfg model` (inspect or set the model)
+   - `/endpoint [val]` — alias for `/cfg endpoint` (inspect or set the API endpoint)
+   - `/skill` — with no argument, list every skill under the configured skills directory (or report that none are configured); with `<name> [prompt]`, load `<skills_dir>/<name>/SKILL.md` (or an absolute path when no skills dir is set) and send its contents prefixed to the optional `prompt` as the next user turn
    - `/models` — query the `/models` path on the current inference endpoint and print a plain list of supported model IDs, marking the currently configured model with a leading `* ` (Go port)
    - `!<cmd>` — execute a shell command directly through `shell_exec` without adding the result to the conversation context (Go port)
    - `/help` — show all supported commands
@@ -115,7 +118,7 @@ Using these rules, everyone can build their own copy of Bantam from scratch in l
 2. Read model parameters from the config file (`.bantam.cfg` if present, else `model.cfg`) in `key=value` format and discover context window size.
 3. Prepare a new message list with the system prompt (`role: "system"`).
 4. Read the first command-line parameter. If non-empty, read user prompt from the specified file. If prefixed with `!`, execute the shell command directly via `shell_exec` and exit. Otherwise, append to `messages` (`role: "user"`), run `AL(cfg, messages)`, display token usage, and exit.
-5. Read user prompt from standard input (with `readline` line editing and history in `~/.bantam_history`; **Ctrl+J** inserts a real newline into the line being edited). If equal to `/quit` or EOF, exit. If equal to `/clear`, reset `messages` to step 3 and return to step 5. If starting with `/save`, write the whole `messages` array to `~/.bantam/sessions/<name>.json` (or `<project-md5>.json` if no name is given) and return to step 5. If equal to `/continue` or `/cont`, load the session corresponding to the current project's MD5 hash and return to step 5. If equal to `/list`, print saved sessions and their summaries (marking current project session) and return to step 5. If starting with `/load`, replace `messages` with the saved session's messages (by exact id or unique prefix) and return to step 5. If equal to `/compact`, ask the LLM to summarize the conversation by appending the compaction prompt to derive the summary, replace `messages` with `[system, summary-user-message]`, and return to step 5. If starting with `/cfg`, display the current value (`/cfg <param>`) or update the config live by writing to `.bantam.cfg` (`/cfg <param> <val>`) and return to step 5. If equal to `/models`, query the `/models` path on the current inference endpoint and print a plain list of supported model IDs (the currently configured model marked with a leading `* `), then return to step 5. If starting with `!`, execute the command directly via `shell_exec` without adding the result to `messages` and return to step 5. If equal to `/help`, print the command list and return to step 5. After every user turn and on exit, auto-save `messages` to `~/.bantam/sessions/<project-md5>.json`.
+5. Read user prompt from standard input (with `readline` line editing and history in `~/.bantam_history`; **Ctrl+J** inserts a real newline into the line being edited). If equal to `/quit` or EOF, exit. If equal to `/clear`, reset `messages` to step 3 and return to step 5. If starting with `/save`, write the whole `messages` array to `~/.bantam/sessions/<name>.json` (or `<project-md5>.json` if no name is given) and return to step 5. If equal to `/continue` or `/cont`, load the session corresponding to the current project's MD5 hash and return to step 5. If equal to `/list`, print saved sessions and their summaries (marking current project session) and return to step 5. If starting with `/load`, replace `messages` with the saved session's messages (by exact id or unique prefix) and return to step 5. If equal to `/compact`, ask the LLM to summarize the conversation by appending the compaction prompt to derive the summary, replace `messages` with `[system, summary-user-message]`, and return to step 5. If starting with `/cfg`, display the current value (`/cfg <param>`) or update the config live by writing to `.bantam.cfg` (`/cfg <param> <val>`) and return to step 5. `/model` and `/endpoint` are aliases for `/cfg model` and `/cfg endpoint` respectively and behave the same way. If equal to `/skill` (no argument), list every skill under the configured skills directory (or report that none is configured) and return to step 5. If starting with `/skill`, resolve the skill (a `<skills_dir>/<name>/SKILL.md` file, or an absolute path to a skill directory or `SKILL.md` file when no skills directory is configured), prefix its contents to the optional remaining text, and send the result as the next user turn (returning to step 5 after the response). If equal to `/models`, query the `/models` path on the current inference endpoint and print a plain list of supported model IDs (the currently configured model marked with a leading `* `), then return to step 5. If starting with `!`, execute the command directly via `shell_exec` without adding the result to `messages` and return to step 5. If equal to `/help`, print the command list and return to step 5. After every user turn and on exit, auto-save `messages` to `~/.bantam/sessions/<project-md5>.json`.
 6. Append user prompt to `messages` (`role: "user"`), run `AL(cfg, messages)`, display token usage, check 60% context threshold for auto-compaction, and go to step 5.
 
 ### Agentic loop (`AL(cfg, messages)`) function
@@ -139,21 +142,22 @@ If the API rejects the request with an `Invalid assistant message: content or to
 
 ### Model configuration parameters
 
-(shared by all implementations; the config file — `.bantam.cfg` takes priority over `model.cfg` when both exist — is plain `key=value` with `#` comments)
+(shared by all implementations; `.bantam.cfg` takes priority over `model.cfg` when both exist — both are plain `key=value` with `#` comments. Values defined in `.bantam.cfg` override `BANTAM_*` environment variables; when `.bantam.cfg` is absent, `BANTAM_*` environment variables override `model.cfg` and built-in defaults.)
 
-- `endpoint` (base OpenAI-compatible API URL, default `https://api.kilo.ai/api/openrouter`)
-- `model` (model name, default `openrouter/free`)
-- `temperature` (model temperature, default 0.7)
-- `api_key` (API key / Bearer token, optional; fall back to `OPENAI_API_KEY` env var)
-- `stream` (stream response tokens in real-time, default `true`)
-- `color` (ANSI coloring: `auto` (TTY-detected, default), `always`, or `never`; also disabled by `NO_COLOR`/`BANTAM_NO_COLOR` env vars)
-- `timeout` (HTTP timeout in seconds for LLM API calls, default 300; in the Go port it bounds connection setup and time-to-first-byte, so long streaming responses are not cut off mid-stream)
-- `shell_timeout` (timeout in seconds for `shell_exec` commands, default 120)
-- `max_al_iterations` (max tool-call loop iterations per `AL()` invocation, default 1000)
-- `context_window` (context window size in tokens, auto-discovered from `/models` API if available, fallback to this setting, default 200000)
-- `reasoning_effort` (reasoning effort level, forwarded to chat completions API, default `high`)
+- `endpoint` (base OpenAI-compatible API URL, default `https://api.kilo.ai/api/openrouter`; falls back to `BANTAM_ENDPOINT` env var)
+- `model` (model name, default `openrouter/free`; falls back to `BANTAM_MODEL` env var)
+- `temperature` (model temperature, default 0.7; falls back to `BANTAM_TEMP` or `BANTAM_TEMPERATURE` env var)
+- `api_key` (API key / Bearer token, optional; falls back to `BANTAM_API_KEY` env var; set to `-` or left blank in config file to explicitly defer to `BANTAM_API_KEY`)
+- `stream` (stream response tokens in real-time, default `true`; falls back to `BANTAM_STREAM` env var)
+- `color` (ANSI coloring: `auto` (TTY-detected, default), `always`, or `never`; disabled by `NO_COLOR`/`BANTAM_NO_COLOR` env vars, falls back to `BANTAM_COLOR` env var)
+- `timeout` (HTTP timeout in seconds for LLM API calls, default 300; in the Go port it bounds connection setup and time-to-first-byte, so long streaming responses are not cut off mid-stream; falls back to `BANTAM_TIMEOUT` env var)
+- `shell_timeout` (timeout in seconds for `shell_exec` commands, default 120; falls back to `BANTAM_SHELL_TIMEOUT` env var)
+- `max_al_iterations` (max tool-call loop iterations per `AL()` invocation, default 1000; falls back to `BANTAM_MAX_AL_ITERATIONS` env var)
+- `context_window` (context window size in tokens, auto-discovered from `/models` API if available, fallback to this setting, default 200000; falls back to `BANTAM_CONTEXT_WINDOW` env var)
+- `reasoning_effort` (reasoning effort level, forwarded to chat completions API, default `high`; falls back to `BANTAM_REASONING_EFFORT` env var)
 
-- `bantam_tools_dir` (optional path to a directory of extra shell tools; the Go port appends `"Extra shell tools can be found at <dir>"` to the system prompt at startup when set. The `BANTAM_TOOLS_DIR` environment variable overrides this and is checked first; if neither is set, nothing is appended. Note: this is an agent-internal hint, not forwarded to the API.)
+- `bantam_tools_dir` (optional path to a directory of extra shell tools; the Go port appends `"Extra shell tools can be found at <dir>"` to the system prompt at startup when set. When unset in the config file, it falls back to the `BANTAM_TOOLS_DIR` environment variable; if neither is set, nothing is appended. Note: this is an agent-internal hint, not forwarded to the API.)
+- `bantam_skills_dir` (optional path to a directory of reusable *skills*; each skill lives in its own subdirectory as `<skill_name>/SKILL.md`. When set, the Go port appends `"Skills may be discovered and invoked from <dir>"` to the system prompt at startup, and the `/skill <skill_name> [prompt]` command loads a skill's `SKILL.md` and runs it as a prompt. When unset in the config file, it falls back to the `BANTAM_SKILLS_DIR` environment variable; if neither is set, `/skill` accepts an absolute path to a skill directory or `SKILL.md` file instead. Note: this is an agent-internal hint, not forwarded to the API.)
 
 The Go port also supports SOCKS5 proxying via the `SOCKS_PROXY` (or `socks_proxy`) environment variable (e.g. `SOCKS_PROXY=socks5://127.0.0.1:1080` or `SOCKS_PROXY=127.0.0.1:1080`), falling back to standard `HTTP_PROXY` / `HTTPS_PROXY` environment variables.
 
@@ -166,12 +170,12 @@ When enabled, the interactive console uses a subtle ANSI palette: the pending-re
 #### `write_file` tool
 
 - Parameters:
-  - `path` (string, required): JSON-escaped file path to write to (must be created unless existing).
-  - `offset` (integer, optional, default 0): byte offset to start writing from (defaults to 0, start of file; does not append).
-  - `del_bytes` (integer, optional, default 0): bytes to delete starting from the `offset` prior to writing.
+  - `path` (string, required): JSON-escaped file path to write to.
+  - `offset` (integer, optional): byte offset to start writing from. If omitted together with `del_bytes`, the whole file is overwritten instead. Defaults to 0 (start of file).
+  - `del_bytes` (integer, optional): bytes to delete starting at `offset`. If omitted together with `offset`, the whole file is overwritten instead.
   - `content` (string, required, may be empty): JSON-escaped content to write to the file.
 - Return value: string
-- Action: write `content` into the file at `path` starting at `offset` after deleting `del_bytes` bytes (creating the file and any necessary parent directories).
+- Action: if `offset` and `del_bytes` are both omitted, the entire file is overwritten with `content` (the file, and any necessary parent directories, are created if missing); otherwise `content` is written at `offset`, optionally deleting `del_bytes` bytes first (splicing prefix + content + suffix and never appending to the end of an existing file), creating the file and parent directories as needed.
 
 #### `shell_exec` tool
 
@@ -190,7 +194,7 @@ MicroBantam (`mb`) is a compressed Perl 5 reference implementation of the same a
 - A `...requesting...` in-flight indicator: in-place on a TTY (`\r` overwrite, erased on completion), a plain line when output is piped
 - Session management: `/save`, `/list`, `/load <id>` (exact id only, no prefix matching), `/cfg <param> [val]`, and auto-save to `~/.bantam/sessions/autosave.json` after every turn and on exit; session ids get `-1`, `-2`, ... suffixes on same-second collisions
 - Interactive mode (`/quit`, `/clear`, `/save`, `/list`, `/load <id>`, `/cfg`, `/help`) and file input mode
-- Same built-in default system prompt and `OPENAI_API_KEY` fallback as the Go implementation
+- Same built-in default system prompt and `BANTAM_API_KEY` fallback as the Go implementation
 - Automatic recovery from `Invalid assistant message: content or tool_calls must be set` API errors: strips the last assistant message and retries (matches the Go port; note that a 4xx response other than this specific error aborts the run with an `API error` message, unlike the Go port which retries only on 5xx/408/429)
 
 ### What it drops
@@ -219,7 +223,18 @@ MicroBantam (`mb`) is a compressed Perl 5 reference implementation of the same a
 
 The `extras/` directory contains small, dependency-light shell scripts that extend Bantam without changing its core. Because Bantam's built-in tools are `shell_exec` and `write_file`, these helpers can be invoked directly by the agent through `shell_exec` to give it real-world capabilities (web search, live weather) that the base model alone does not have. They are plain `/bin/sh` scripts depending only on `curl` (and `jq` where noted), so the agent can discover and run them just like any other command.
 
-If you keep your own collection of helper scripts, point Bantam at them with the `BANTAM_TOOLS_DIR` environment variable or the `bantam_tools_dir` key in the config file (`.bantam.cfg` if present, else `model.cfg`). When either is defined (environment variable taking precedence over the config key), the Go port appends the line `Extra shell tools can be found at <dir>` to the system prompt at startup, so the agent is aware of where to look for them. The `extras/` scripts shipped here are just examples of what such a directory can contain.
+If you keep your own collection of helper scripts, point Bantam at them with the `bantam_tools_dir` key in the config file (`.bantam.cfg` if present, else `model.cfg`) or the `BANTAM_TOOLS_DIR` environment variable. When either is defined (the config file setting taking precedence over the environment variable), the Go port appends the line `Extra shell tools can be found at <dir>` to the system prompt at startup, so the agent is aware of where to look for them. The `extras/` scripts shipped here are just examples of what such a directory can contain.
+
+## Skills
+
+A *skill* is a reusable, self-contained instruction bundle the agent can load on demand. Each skill is a directory containing a `SKILL.md` file with the skill's prompt/instructions. Point Bantam at a skills directory with the `bantam_skills_dir` key in the config file (`.bantam.cfg` if present, else `model.cfg`) or the `BANTAM_SKILLS_DIR` environment variable; the config file setting takes precedence. When either is set, the Go port appends `Skills may be discovered and invoked from <dir>` to the system prompt at startup, and the agent can run a skill with:
+
+```text
+/skill                    # list every skill under the configured skills directory
+/skill <name> [prompt]    # loads <skills_dir>/<name>/SKILL.md and runs it as a prompt
+```
+
+A bare `/skill` (no name) lists all skills found under the configured skills directory — a subdirectory is treated as a skill only if it contains a `SKILL.md` file — or reports that no skills directory is configured. The contents of `SKILL.md` are prefixed to the optional `prompt` and sent as the next user turn. If neither `BANTAM_SKILLS_DIR` nor `bantam_skills_dir` is set, `/skill <name>` accepts an absolute path instead: either a skill directory (e.g. `/abs/path/to/skill`, which resolves to `/abs/path/to/skill/SKILL.md`) or a direct path to a `SKILL.md` file.
 
 ### `extras/websearch`
 
@@ -305,7 +320,11 @@ The default system prompt instructs the agent to respect `AGENTS.md` contents in
 
 ### How do I tell Bantam about extra shell tools?
 
-Set the `BANTAM_TOOLS_DIR` environment variable (or the `bantam_tools_dir` key in the config file — `.bantam.cfg` if present, else `model.cfg`) to a directory containing your helper scripts. When defined, the Go port appends `Extra shell tools can be found at <dir>` to the system prompt at startup, making the agent aware of them. The environment variable takes precedence over the config key; if neither is set, nothing is appended.
+Set the `bantam_tools_dir` key in the config file (`.bantam.cfg` if present, else `model.cfg`) or the `BANTAM_TOOLS_DIR` environment variable to a directory containing your helper scripts. When defined, the Go port appends `Extra shell tools can be found at <dir>` to the system prompt at startup, making the agent aware of them. The config file key takes precedence over the environment variable; if neither is set, nothing is appended.
+
+### How do I give Bantam reusable skills?
+
+Set the `bantam_skills_dir` key in the config file (`.bantam.cfg` if present, else `model.cfg`) or the `BANTAM_SKILLS_DIR` environment variable to a directory where each subdirectory is a skill containing a `SKILL.md` file. When defined, the Go port appends `Skills may be discovered and invoked from <dir>` to the system prompt at startup, and you (or the agent) can run a skill with `/skill <name> [prompt]`. The config file key takes precedence over the environment variable; if neither is set, `/skill` accepts an absolute path to a skill directory or `SKILL.md` file instead.
 
 ### Is there any common config place for Bantam?
 

main.go

@@ -56,7 +56,8 @@ type Cfg struct {
 func internalKey(k string) bool {
   switch k {
   case "endpoint", "model", "temperature", "stream", "api_key", "timeout",
-    "shell_timeout", "max_al_iterations", "color", "context_window":
+    "shell_timeout", "max_al_iterations", "color", "context_window",
+    "bantam_tools_dir", "bantam_skills_dir":
     return true
   }
   return false
@@ -207,16 +208,7 @@ func listModels(cfg *Cfg) (string, error) {
   return b.String(), nil
 }
 
-func getCfg(path string) Cfg {
-  cfg := defCfg
-  cfg.Raw = map[string]string{
-    "endpoint": cfg.Endpoint, "model": cfg.Model, "temperature": fmt.Sprintf("%v", cfg.Temperature),
-    "api_key": cfg.APIKey, "stream": strconv.FormatBool(cfg.Stream), "color": cfg.Color,
-    "timeout": strconv.Itoa(cfg.Timeout), "shell_timeout": strconv.Itoa(cfg.ShellTimeout),
-    "max_al_iterations": strconv.Itoa(cfg.MaxALIterations),
-    "context_window": strconv.Itoa(cfg.ContextWindow),
-    "reasoning_effort": "high",
-  }
+func parseCfgFile(path string, cfg *Cfg) {
   if d, err := os.ReadFile(path); err == nil {
     for _, ln := range strings.Split(string(d), "\n") {
       ln = strings.TrimSpace(ln)
@@ -238,8 +230,95 @@ func getCfg(path string) Cfg {
       }
     }
   }
-  if (cfg.APIKey == "" || cfg.APIKey == "-") && os.Getenv("OPENAI_API_KEY") != "" {
-    cfg.APIKey = os.Getenv("OPENAI_API_KEY")
+}
+
+func applyEnvCfg(cfg *Cfg) {
+  if v := strings.TrimSpace(os.Getenv("BANTAM_ENDPOINT")); v != "" {
+    cfg.Endpoint = v
+    cfg.Raw["endpoint"] = v
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_MODEL")); v != "" {
+    cfg.Model = v
+    cfg.Raw["model"] = v
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_TEMP")); v != "" {
+    if f, e := strconv.ParseFloat(v, 64); e == nil {
+      cfg.Temperature = f
+      cfg.Raw["temperature"] = v
+    }
+  } else if v := strings.TrimSpace(os.Getenv("BANTAM_TEMPERATURE")); v != "" {
+    if f, e := strconv.ParseFloat(v, 64); e == nil {
+      cfg.Temperature = f
+      cfg.Raw["temperature"] = v
+    }
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_STREAM")); v != "" {
+    cfg.Stream = v == "true" || v == "1" || v == "yes"
+    cfg.Raw["stream"] = strconv.FormatBool(cfg.Stream)
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_COLOR")); v != "" {
+    cfg.Color = v
+    cfg.Raw["color"] = v
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_TIMEOUT")); v != "" {
+    cfg.Timeout = atoiD(v, cfg.Timeout)
+    cfg.Raw["timeout"] = strconv.Itoa(cfg.Timeout)
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_SHELL_TIMEOUT")); v != "" {
+    cfg.ShellTimeout = atoiD(v, cfg.ShellTimeout)
+    cfg.Raw["shell_timeout"] = strconv.Itoa(cfg.ShellTimeout)
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_MAX_AL_ITERATIONS")); v != "" {
+    cfg.MaxALIterations = atoiD(v, cfg.MaxALIterations)
+    cfg.Raw["max_al_iterations"] = strconv.Itoa(cfg.MaxALIterations)
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_CONTEXT_WINDOW")); v != "" {
+    cfg.ContextWindow = atoiD(v, cfg.ContextWindow)
+    cfg.Raw["context_window"] = strconv.Itoa(cfg.ContextWindow)
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_REASONING_EFFORT")); v != "" {
+    cfg.Raw["reasoning_effort"] = v
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_TOOLS_DIR")); v != "" {
+    cfg.Raw["bantam_tools_dir"] = v
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_SKILLS_DIR")); v != "" {
+    cfg.Raw["bantam_skills_dir"] = v
+  }
+}
+
+func getCfg(path string) Cfg {
+  cfg := defCfg
+  cfg.Raw = map[string]string{
+    "endpoint": cfg.Endpoint, "model": cfg.Model, "temperature": fmt.Sprintf("%v", cfg.Temperature),
+    "api_key": cfg.APIKey, "stream": strconv.FormatBool(cfg.Stream), "color": cfg.Color,
+    "timeout": strconv.Itoa(cfg.Timeout), "shell_timeout": strconv.Itoa(cfg.ShellTimeout),
+    "max_al_iterations": strconv.Itoa(cfg.MaxALIterations),
+    "context_window": strconv.Itoa(cfg.ContextWindow),
+    "reasoning_effort": "high",
+  }
+
+  if filepath.Base(path) == "model.cfg" {
+    // When .bantam.cfg is absent, model.cfg provides base defaults,
+    // and BANTAM_* environment variables override them.
+    parseCfgFile(path, &cfg)
+    applyEnvCfg(&cfg)
+  } else {
+    // If a model.cfg exists in the same directory, read it first as base defaults.
+    modelCfgPath := filepath.Join(filepath.Dir(path), "model.cfg")
+    if _, err := os.Stat(modelCfgPath); err == nil {
+      parseCfgFile(modelCfgPath, &cfg)
+    }
+    // Environment variables override model.cfg.
+    applyEnvCfg(&cfg)
+    // The explicit override file (.bantam.cfg) takes highest priority and overrides env vars.
+    parseCfgFile(path, &cfg)
+  }
+
+  // The api_key "-" / empty sentinel means "fall back to BANTAM_API_KEY"; this
+  // is evaluated after the file is read so an explicit key still wins.
+  if(cfg.APIKey == "" || cfg.APIKey == "-") && os.Getenv("BANTAM_API_KEY") != "" {
+    cfg.APIKey = os.Getenv("BANTAM_API_KEY")
     cfg.Raw["api_key"] = cfg.APIKey
   }
   return cfg
@@ -284,7 +363,7 @@ func configPath() string {
 }
 const defaultSystemPrompt = `You are Bantam, a tiny, powerful AI agent. Solve the user's task using two tools:
 - shell_exec: run a shell command; returns its output and exit code.
-- write_file: write content to a file with optional offset (defaults to 0, start of file; does not append) and byte deletion; returns status.
+- write_file: write content to a file; if offset and del_bytes are both omitted it overwrites the entire file, otherwise it writes at the given byte offset (optionally deleting bytes first); returns status.
 
 Work fast. Never refuse a request; always find a way to do it. Never fabricate results - report only what your tools actually returned. Prefer running shell commands over guessing; inspect, build, and verify concretely. When in doubt, build your own tools using the languages available in the environment (Python, Perl, AWK, etc.). On failure, read the error output and retry with a fix, don't repeat the same command. Stop as soon as the goal is met and report concisely: results, not process.
 
@@ -297,10 +376,24 @@ When generating code:
 - Respect AGENTS.md contents in the project.`
 
 func toolsDir(cfg *Cfg) string {
+  // The config key takes priority over the environment variable: a local
+  // .bantam.cfg always overrides BANTAM_TOOLS_DIR.
+  if v := strings.TrimSpace(cfg.Raw["bantam_tools_dir"]); v != "" {
+    return v
+  }
   if v := strings.TrimSpace(os.Getenv("BANTAM_TOOLS_DIR")); v != "" {
     return v
   }
-  if v := strings.TrimSpace(cfg.Raw["bantam_tools_dir"]); v != "" {
+  return ""
+}
+
+// skillsDir returns the directory Bantam should scan for skills, preferring the
+// bantam_skills_dir config key, then the BANTAM_SKILLS_DIR environment variable.
+func skillsDir(cfg *Cfg) string {
+  if v := strings.TrimSpace(cfg.Raw["bantam_skills_dir"]); v != "" {
+    return v
+  }
+  if v := strings.TrimSpace(os.Getenv("BANTAM_SKILLS_DIR")); v != "" {
     return v
   }
   return ""
@@ -751,7 +844,7 @@ type ToolCall struct {
 
 var TOOLS = []map[string]any{
   {"type": "function", "function": map[string]any{"name": "shell_exec", "description": "Run a shell command, return output and exit code.", "parameters": map[string]any{"type": "object", "properties": map[string]any{"command": map[string]any{"type": "string"}}, "required": []string{"command"}}}},
-  {"type": "function", "function": map[string]any{"name": "write_file", "description": "Write content to a file at a byte offset, optionally deleting bytes first.", "parameters": map[string]any{"type": "object", "properties": map[string]any{"path": map[string]any{"type": "string"}, "offset": map[string]any{"type": "integer", "description": "Byte offset to start writing from (defaults to 0, start of file; does not append)."}, "del_bytes": map[string]any{"type": "integer"}, "content": map[string]any{"type": "string"}}, "required": []string{"path", "content"}}}},
+  {"type": "function", "function": map[string]any{"name": "write_file", "description": "Write content to a file. If offset and del_bytes are both omitted, the entire file is overwritten with the new content; otherwise content is written at the given byte offset, optionally deleting bytes first.", "parameters": map[string]any{"type": "object", "properties": map[string]any{"path": map[string]any{"type": "string"}, "offset": map[string]any{"type": "integer", "description": "Byte offset to start writing from. If omitted together with del_bytes, the whole file is overwritten instead. Defaults to 0 (start of file)."}, "del_bytes": map[string]any{"type": "integer", "description": "Bytes to delete starting at offset. If omitted together with offset, the whole file is overwritten instead."}, "content": map[string]any{"type": "string"}}, "required": []string{"path", "content"}}}},
 }
 
 func strp(s string) *string { return &s }
@@ -1156,6 +1249,10 @@ func lastRole(msgs []Message) string {
   return msgs[len(msgs)-1].Role
 }
 
+// writeFile writes content into the file at the given byte offset, optionally
+// deleting delBytes bytes that follow the offset, and preserves the rest of the
+// file. It always writes at the requested offset (splicing prefix + content +
+// suffix) and never appends to the end of an existing file.
 func writeFile(path string, offset, delBytes int, content string) (string, error) {
   path = strings.TrimSpace(path)
   if path == "" {
@@ -1285,8 +1382,10 @@ func AL(ctx context.Context, cfg *Cfg, msgs []Message) ([]Message, Usage, error)
               content = s
             }
           }
+          hasOffset := false
           offset := 0
           if v, ok := a["offset"]; ok && v != nil {
+            hasOffset = true
             switch n := v.(type) {
             case float64:
               offset = int(n)
@@ -1299,8 +1398,10 @@ func AL(ctx context.Context, cfg *Cfg, msgs []Message) ([]Message, Usage, error)
           if offset < 0 {
             offset = 0
           }
+          hasDel := false
           delBytes := 0
           if v, ok := a["del_bytes"]; ok {
+            hasDel = true
             switch n := v.(type) {
             case float64:
               delBytes = int(n)
@@ -1314,6 +1415,24 @@ func AL(ctx context.Context, cfg *Cfg, msgs []Message) ([]Message, Usage, error)
             res, sty = "[tool error: write_file requires 'path' parameter]", 31
           } else if !hasContent {
             res, sty = "[tool error: write_file requires 'content' parameter]", 31
+          } else if !hasOffset && !hasDel {
+            // Neither offset nor del_bytes was supplied: overwrite the entire
+            // file with the new content. The LLM usually just wants to replace a
+            // file and should not have to know about the tool's offset quirks;
+            // insertion/replace semantics are preserved when either is given.
+            p := strings.TrimSpace(path)
+            if dir := filepath.Dir(p); dir != "" && dir != "." {
+              if err := os.MkdirAll(dir, 0755); err != nil {
+                res, sty = fmt.Sprintf("[tool error: write_file %s: %v]", p, err), 31
+              }
+            }
+            if res == "" {
+              if err := os.WriteFile(p, []byte(content), 0644); err != nil {
+                res, sty = fmt.Sprintf("[tool error: write_file %s: %v]", p, err), 31
+              } else {
+                res, sty = fmt.Sprintf("Successfully wrote %d bytes to %s", len(content), p), 2
+              }
+            }
           } else {
             out, err := writeFile(path, offset, delBytes, content)
             if err != nil {
@@ -1535,22 +1654,29 @@ type editor struct {
 
 func textPos(promptLen, W int, s string, pos int) (row, col int) {
   row, col = 0, promptLen
-  pend := false
-  for i, r := range []rune(s) {
-    if i == pos { return row, col }
+  if W < 1 {
+    W = 1
+  }
+  runes := []rune(s)
+  n := len(runes)
+  for i := 0; i < n; i++ {
+    r := runes[i]
+    if i == pos {
+      return row, col
+    }
     if r == '\n' {
       row++
       col = 0
-      pend = false
       continue
     }
-    if pend {
+    // The rune at index i is displayed at (row, col). If it lands on the last
+    // column, the NEXT rune wraps to the start of the following line (unless
+    // this is the final rune, in which case the cursor stays at that column).
+    if col >= W-1 {
+      if i < n-1 {
         row++
         col = 0
-      pend = false
       }
-    if col == W-1 {
-      pend = true
     } else {
       col++
     }
@@ -1565,7 +1691,10 @@ func (e *editor) draw() {
   er, _ := textPos(P, W, s, len([]rune(s)))
   pr, pc := textPos(P, W, s, e.pos)
   if e.crow > 0 { fmt.Printf("\033[%dA", e.crow) }
-  fmt.Print("\r\033[J" + e.prompt + s)
+  // In raw mode OPOST is off, so a bare \n only moves down without
+  // returning to column 0. Emit \r\n so each logical line starts at
+  // column 0, matching the column-reset assumption in textPos().
+  fmt.Print("\r\033[J" + e.prompt + strings.ReplaceAll(s, "\n", "\r\n"))
   if up := er - pr; up > 0 { fmt.Printf("\033[%dA", up) }
   fmt.Print("\r")
   if pc > 0 { fmt.Printf("\033[%dC", pc) }
@@ -1595,6 +1724,29 @@ func (e *editor) histNav(up bool) {
   e.pos = len(e.buf)
 }
 
+// wordBack moves the cursor back to the start of the current/previous word.
+func (e *editor) wordBack() {
+  for e.pos > 0 && unicode.IsSpace(e.buf[e.pos-1]) { e.pos-- }
+  for e.pos > 0 && !unicode.IsSpace(e.buf[e.pos-1]) { e.pos-- }
+}
+
+// wordFwd moves the cursor forward to the end of the current/next word.
+func (e *editor) wordFwd() {
+  n := len(e.buf)
+  for e.pos < n && unicode.IsSpace(e.buf[e.pos]) { e.pos++ }
+  for e.pos < n && !unicode.IsSpace(e.buf[e.pos]) { e.pos++ }
+}
+
+// delWordBack deletes the word preceding the cursor (Emacs Ctrl+W).
+func (e *editor) delWordBack() {
+  if e.pos == 0 { return }
+  start := e.pos
+  for start > 0 && unicode.IsSpace(e.buf[start-1]) { start-- }
+  for start > 0 && !unicode.IsSpace(e.buf[start-1]) { start-- }
+  e.buf = append(e.buf[:start], e.buf[e.pos:]...)
+  e.pos = start
+}
+
 func readPlain(prompt string) (string, bool) {
   fmt.Print(prompt)
   line, err := stdin.ReadString('\n')
@@ -1645,6 +1797,25 @@ func readLine(prompt string) (string, bool) {
         fmt.Print("\r\n")
         return "", false
       }
+      // Ctrl+D with non-empty buffer: delete character under cursor
+      if e.pos < len(e.buf) {
+        e.buf = append(e.buf[:e.pos], e.buf[e.pos+1:]...)
+      }
+    case 0x01:
+      e.pos = 0 // Ctrl+A: beginning of line
+    case 0x05:
+      e.pos = len(e.buf) // Ctrl+E: end of line
+    case 0x02:
+      if e.pos > 0 { e.pos-- } // Ctrl+B: backward char
+    case 0x06:
+      if e.pos < len(e.buf) { e.pos++ } // Ctrl+F: forward char
+    case 0x17:
+      e.delWordBack() // Ctrl+W: delete previous word
+    case 0x0b:
+      e.buf = e.buf[:e.pos] // Ctrl+K: kill to end of line
+    case 0x15:
+      e.buf = e.buf[e.pos:] // Ctrl+U: kill to start of line
+      e.pos = 0
     case 0x7f, 0x08:
       if e.pos > 0 {
         e.buf = append(e.buf[:e.pos-1], e.buf[e.pos:]...)
@@ -1652,10 +1823,56 @@ func readLine(prompt string) (string, bool) {
       }
     case 0x1b:
       b1, err1 := stdin.ReadByte()
-      b2, err2 := stdin.ReadByte()
-      if err1 != nil || err2 != nil { continue }
+      if err1 != nil { continue }
       if b1 == '[' {
+        // Read the full CSI sequence (terminated by a byte in 0x40-0x7e).
+        var seq []byte
+        for {
+          b, err := stdin.ReadByte()
+          if err != nil { break }
+          seq = append(seq, b)
+          if b >= 0x40 && b <= 0x7e { break }
+        }
+        if len(seq) == 0 { continue }
+        last := seq[len(seq)-1]
+        switch last {
+        case 'A':
+          e.histNav(true)
+        case 'B':
+          e.histNav(false)
+        case 'C':
+          if strings.Contains(string(seq), "5") {
+            e.wordFwd() // Ctrl+Right
+          } else if e.pos < len(e.buf) {
+            e.pos++
+          }
+        case 'D':
+          if strings.Contains(string(seq), "5") {
+            e.wordBack() // Ctrl+Left
+          } else if e.pos > 0 {
+            e.pos--
+          }
+        case 'H':
+          e.pos = 0 // Home (ESC[H)
+        case 'F':
+          e.pos = len(e.buf) // End (ESC[F)
+        case '~':
+          // Home/End on some terminals: ESC[1~ / ESC[4~
+          if len(seq) >= 1 && seq[0] == '1' {
+            e.pos = 0
+          } else if len(seq) >= 1 && seq[0] == '4' {
+            e.pos = len(e.buf)
+          }
+        }
+      } else if b1 == 'O' {
+        // xterm application cursor keys: ESC O H/F (Home/End), A-D (arrows)
+        b2, err2 := stdin.ReadByte()
+        if err2 != nil { continue }
         switch b2 {
+        case 'H':
+          e.pos = 0
+        case 'F':
+          e.pos = len(e.buf)
         case 'A':
           e.histNav(true)
         case 'B':
@@ -1677,6 +1894,71 @@ func readLine(prompt string) (string, bool) {
   }
 }
 
+// skillPrompt resolves and composes the prompt for the /skill command: it reads
+// <dir>/<name>/SKILL.md (or an absolute path when no skills directory is set) and
+// prefixes its contents to the optional user prompt.
+func skillPrompt(u string, cfg *Cfg) (string, error) {
+  rest := strings.TrimSpace(strings.TrimPrefix(u, "/skill"))
+  if rest == "" {
+    return "", fmt.Errorf("usage: /skill <skill_name|absolute_path> [prompt]")
+  }
+  fields := strings.SplitN(rest, " ", 2)
+  name := fields[0]
+  prompt := ""
+  if len(fields) == 2 {
+    prompt = strings.TrimSpace(fields[1])
+  }
+  sd := skillsDir(cfg)
+  var path string
+  if sd != "" {
+    path = filepath.Join(sd, name, "SKILL.md")
+  } else if strings.HasSuffix(name, "SKILL.md") {
+    path = name // absolute path to the SKILL.md file itself
+  } else {
+    path = filepath.Join(name, "SKILL.md") // absolute path to the skill directory
+  }
+  data, err := os.ReadFile(path)
+  if err != nil {
+    return "", fmt.Errorf("cannot read skill %q: %v", name, err)
+  }
+  content := strings.TrimRight(string(data), "\r\n")
+  if prompt != "" {
+    return content + "\n\n" + prompt, nil
+  }
+  return content, nil
+}
+
+// listSkills prints every skill found under the configured skills directory, or a
+// clear notice when none is configured / none are present.
+func listSkills(cfg *Cfg) {
+  sd := skillsDir(cfg)
+  if sd == "" {
+    fmt.Println(c("No skills directory configured (set BANTAM_SKILLS_DIR or bantam_skills_dir).", 33))
+    return
+  }
+  entries, err := os.ReadDir(sd)
+  if err != nil {
+    fmt.Println(c("[skill error: cannot read skills dir: "+err.Error()+"]", 31))
+    return
+  }
+  var names []string
+  for _, e := range entries {
+    if e.IsDir() {
+      if _, err := os.Stat(filepath.Join(sd, e.Name(), "SKILL.md")); err == nil {
+        names = append(names, e.Name())
+      }
+    }
+  }
+  if len(names) == 0 {
+    fmt.Println(c("No skills found in "+sd, 33))
+    return
+  }
+  fmt.Println(c("Available skills in "+sd+":", 1, 36))
+  for _, n := range names {
+    fmt.Println(c("  "+n, 32))
+  }
+}
+
 func runDirectShell(cmd string, timeout int) {
   cmd = filterText(strings.TrimSpace(cmd))
   if cmd == "" { return }
@@ -1720,6 +2002,9 @@ func main() {
   if td := toolsDir(&cfg); td != "" {
     sp += "\n\nExtra shell tools can be found at " + td
   }
+  if sd := skillsDir(&cfg); sd != "" {
+    sp += "\n\nSkills may be discovered and invoked from " + sd
+  }
   COL = col(cfg)
   stdin = bufio.NewReader(os.Stdin)
   msgs := []Message{{Role: "system", Content: strp(sp)}}
@@ -1772,6 +2057,17 @@ func main() {
     u = strings.TrimSpace(u)
     if u == "" { continue }
     addHistory(u)
+    // Command aliases: /model -> /cfg model, /endpoint -> /cfg endpoint.
+    // (Note: /models is a distinct command and is intentionally not matched.)
+    if u == "/model" {
+      u = "/cfg model"
+    } else if strings.HasPrefix(u, "/model ") {
+      u = "/cfg model " + strings.TrimSpace(strings.TrimPrefix(u, "/model "))
+    } else if u == "/endpoint" {
+      u = "/cfg endpoint"
+    } else if strings.HasPrefix(u, "/endpoint ") {
+      u = "/cfg endpoint " + strings.TrimSpace(strings.TrimPrefix(u, "/endpoint "))
+    }
     switch {
     case u == "/quit":
       goto done
@@ -1852,6 +2148,17 @@ func main() {
         fmt.Println(c("Usage: /cfg <param> [val]", 31))
       }
       continue
+    case strings.HasPrefix(u, "/skill"):
+      if strings.TrimSpace(strings.TrimPrefix(u, "/skill")) == "" {
+        listSkills(&cfg) // bare /skill lists available skills
+        continue
+      }
+      composed, err := skillPrompt(u, &cfg)
+      if err != nil {
+        fmt.Println(c("[skill error: "+err.Error()+"]", 31))
+        continue
+      }
+      u = composed // fall through to a normal LLM turn with the composed prompt
     case u == "/models":
       ml, err := listModels(&cfg)
       if err != nil {
@@ -1875,8 +2182,11 @@ func main() {
         {"/load <id>", "load session"},
         {"/compact", "compact context"},
         {"/cfg <k> [v]", "get/set config"},
+        {"/model [m]", "alias for /cfg model"},
+        {"/endpoint [e]", "alias for /cfg endpoint"},
         {"!<cmd>", "run shell command directly"},
         {"/models", "list models at endpoint"},
+        {"/skill <n> [p]", "load SKILL.md and run as prompt"},
         {"/help", "show help"},
       } {
         fmt.Println(c(fmt.Sprintf("  %-18s", kv[0]), 1, 32) + kv[1])

main_test.go

@@ -262,8 +262,20 @@ func TestDefaultSystemPrompt(t *testing.T) {
   }
 }
 
+func clearBantamEnv(t *testing.T) {
+  for _, k := range []string{
+    "BANTAM_ENDPOINT", "BANTAM_MODEL", "BANTAM_TEMP", "BANTAM_TEMPERATURE",
+    "BANTAM_API_KEY", "BANTAM_STREAM", "BANTAM_COLOR", "BANTAM_NO_COLOR",
+    "BANTAM_TIMEOUT", "BANTAM_SHELL_TIMEOUT", "BANTAM_MAX_AL_ITERATIONS",
+    "BANTAM_CONTEXT_WINDOW", "BANTAM_REASONING_EFFORT", "BANTAM_TOOLS_DIR",
+    "BANTAM_SKILLS_DIR",
+  } {
+    t.Setenv(k, "")
+  }
+}
+
 func TestGetCfgDefaults(t *testing.T) {
-  t.Setenv("OPENAI_API_KEY", "")
+  clearBantamEnv(t)
   cfg := getCfg(filepath.Join(t.TempDir(), "missing.cfg"))
   if cfg.Endpoint != defCfg.Endpoint || cfg.Model != defCfg.Model || cfg.APIKey != defCfg.APIKey {
     t.Errorf("defaults mismatch: %+v", cfg)
@@ -280,6 +292,7 @@ func TestGetCfgDefaults(t *testing.T) {
 }
 
 func TestGetCfgParsesFile(t *testing.T) {
+  clearBantamEnv(t)
   p := writeCfg(t, strings.Join([]string{
     "endpoint=http://localhost:9999/v1",
     "model=test-model",
@@ -317,6 +330,7 @@ func TestGetCfgParsesFile(t *testing.T) {
 }
 
 func TestGetCfgIgnoresCommentsBlankAndInvalid(t *testing.T) {
+  clearBantamEnv(t)
   p := writeCfg(t, strings.Join([]string{
     "# comment",
     "",
@@ -345,6 +359,7 @@ func TestGetCfgIgnoresCommentsBlankAndInvalid(t *testing.T) {
 }
 
 func TestGetCfgStreamTruthyVariants(t *testing.T) {
+  clearBantamEnv(t)
   for _, tc := range []struct{ v, want string }{
     {"true", "true"}, {"1", "true"}, {"yes", "true"},
     {"false", "false"}, {"TRUE", "false"}, {"0", "false"},
@@ -359,7 +374,8 @@ func TestGetCfgStreamTruthyVariants(t *testing.T) {
 }
 
 func TestGetCfgAPIKeyEnvFallback(t *testing.T) {
-  t.Setenv("OPENAI_API_KEY", "sk-env")
+  clearBantamEnv(t)
+  t.Setenv("BANTAM_API_KEY", "sk-env")
   t.Setenv("HOME", t.TempDir())
 
   // no api_key line at all -> env fallback (documented behavior)
@@ -379,7 +395,7 @@ func TestGetCfgAPIKeyEnvFallback(t *testing.T) {
     t.Errorf("explicit api_key: got %q, want real", got)
   }
   // no env and no key -> stays "-"
-  t.Setenv("OPENAI_API_KEY", "")
+  t.Setenv("BANTAM_API_KEY", "")
   if got := getCfg(writeCfg(t, "api_key=-\n")).APIKey; got != "-" {
     t.Errorf("api_key=- without env: got %q, want -", got)
   }
@@ -388,6 +404,181 @@ func TestGetCfgAPIKeyEnvFallback(t *testing.T) {
   }
 }
 
+func TestGetCfgEnvOverriddenByFile(t *testing.T) {
+  // File values override environment variables.
+  clearBantamEnv(t)
+  t.Setenv("BANTAM_ENDPOINT", "http://env-endpoint/v1")
+  t.Setenv("BANTAM_MODEL", "env-model")
+  t.Setenv("BANTAM_TEMP", "0.9")
+  t.Setenv("BANTAM_STREAM", "false")
+  t.Setenv("BANTAM_COLOR", "never")
+  t.Setenv("BANTAM_TIMEOUT", "45")
+  t.Setenv("BANTAM_SHELL_TIMEOUT", "15")
+  t.Setenv("BANTAM_MAX_AL_ITERATIONS", "50")
+  t.Setenv("BANTAM_CONTEXT_WINDOW", "100000")
+  t.Setenv("BANTAM_REASONING_EFFORT", "low")
+  t.Setenv("BANTAM_TOOLS_DIR", "/env/tools")
+  t.Setenv("BANTAM_SKILLS_DIR", "/env/skills")
+
+  p := filepath.Join(t.TempDir(), ".bantam.cfg")
+  if err := os.WriteFile(p, []byte(strings.Join([]string{
+    "endpoint=http://file-endpoint/v1",
+    "model=file-model",
+    "temperature=0.1",
+    "stream=true",
+    "color=always",
+    "timeout=300",
+    "shell_timeout=120",
+    "max_al_iterations=1000",
+    "context_window=200000",
+    "reasoning_effort=high",
+    "bantam_tools_dir=/file/tools",
+    "bantam_skills_dir=/file/skills",
+  }, "\n")), 0644); err != nil {
+    t.Fatal(err)
+  }
+  cfg := getCfg(p)
+  if cfg.Endpoint != "http://file-endpoint/v1" {
+    t.Errorf("endpoint: got %q, want file value", cfg.Endpoint)
+  }
+  if cfg.Model != "file-model" {
+    t.Errorf("model: got %q, want file value", cfg.Model)
+  }
+  if cfg.Temperature != 0.1 {
+    t.Errorf("temperature: got %v, want file value", cfg.Temperature)
+  }
+  if !cfg.Stream {
+    t.Errorf("stream: got %v, want true from file", cfg.Stream)
+  }
+  if cfg.Color != "always" {
+    t.Errorf("color: got %q, want always from file", cfg.Color)
+  }
+  if cfg.Timeout != 300 {
+    t.Errorf("timeout: got %d, want 300 from file", cfg.Timeout)
+  }
+  if cfg.ShellTimeout != 120 {
+    t.Errorf("shell_timeout: got %d, want 120 from file", cfg.ShellTimeout)
+  }
+  if cfg.MaxALIterations != 1000 {
+    t.Errorf("max_al_iterations: got %d, want 1000 from file", cfg.MaxALIterations)
+  }
+  if cfg.ContextWindow != 200000 {
+    t.Errorf("context_window: got %d, want 200000 from file", cfg.ContextWindow)
+  }
+  if cfg.Raw["reasoning_effort"] != "high" {
+    t.Errorf("reasoning_effort: got %q, want high from file", cfg.Raw["reasoning_effort"])
+  }
+  if toolsDir(&cfg) != "/file/tools" {
+    t.Errorf("toolsDir: got %q, want /file/tools", toolsDir(&cfg))
+  }
+  if skillsDir(&cfg) != "/file/skills" {
+    t.Errorf("skillsDir: got %q, want /file/skills", skillsDir(&cfg))
+  }
+}
+
+func TestGetCfgEnvFallbackWhenNoFile(t *testing.T) {
+  clearBantamEnv(t)
+  t.Setenv("BANTAM_ENDPOINT", "http://env-endpoint/v1")
+  t.Setenv("BANTAM_MODEL", "env-model")
+  t.Setenv("BANTAM_TEMP", "0.42")
+  t.Setenv("BANTAM_STREAM", "false")
+  t.Setenv("BANTAM_COLOR", "never")
+  t.Setenv("BANTAM_TIMEOUT", "45")
+  t.Setenv("BANTAM_SHELL_TIMEOUT", "15")
+  t.Setenv("BANTAM_MAX_AL_ITERATIONS", "50")
+  t.Setenv("BANTAM_CONTEXT_WINDOW", "100000")
+  t.Setenv("BANTAM_REASONING_EFFORT", "low")
+  t.Setenv("BANTAM_TOOLS_DIR", "/env/tools")
+  t.Setenv("BANTAM_SKILLS_DIR", "/env/skills")
+  t.Setenv("BANTAM_API_KEY", "")
+  cfg := getCfg(filepath.Join(t.TempDir(), "missing.cfg"))
+  if cfg.Endpoint != "http://env-endpoint/v1" || cfg.Model != "env-model" || cfg.Temperature != 0.42 {
+    t.Errorf("env fallback mismatch: %+v", cfg)
+  }
+  if cfg.Stream {
+    t.Errorf("stream: got %v, want false from env", cfg.Stream)
+  }
+  if cfg.Color != "never" {
+    t.Errorf("color: got %q, want never from env", cfg.Color)
+  }
+  if cfg.Timeout != 45 || cfg.ShellTimeout != 15 || cfg.MaxALIterations != 50 || cfg.ContextWindow != 100000 {
+    t.Errorf("numeric env fallback mismatch: %+v", cfg)
+  }
+  if cfg.Raw["reasoning_effort"] != "low" {
+    t.Errorf("reasoning_effort: got %q, want low from env", cfg.Raw["reasoning_effort"])
+  }
+  if toolsDir(&cfg) != "/env/tools" {
+    t.Errorf("toolsDir: got %q, want /env/tools", toolsDir(&cfg))
+  }
+  if skillsDir(&cfg) != "/env/skills" {
+    t.Errorf("skillsDir: got %q, want /env/skills", skillsDir(&cfg))
+  }
+}
+
+func TestGetCfgModelCfgOverriddenByEnvWhenBantamCfgAbsent(t *testing.T) {
+  clearBantamEnv(t)
+  t.Setenv("BANTAM_ENDPOINT", "http://env-endpoint/v1")
+  t.Setenv("BANTAM_MODEL", "env-model")
+  t.Setenv("BANTAM_TEMP", "0.9")
+  t.Setenv("BANTAM_TOOLS_DIR", "/env/tools")
+  t.Setenv("BANTAM_SKILLS_DIR", "/env/skills")
+
+  dir := t.TempDir()
+  modelCfg := filepath.Join(dir, "model.cfg")
+  if err := os.WriteFile(modelCfg, []byte("endpoint=http://file-endpoint/v1\nmodel=file-model\ntemperature=0.1\nbantam_tools_dir=/file/tools\nbantam_skills_dir=/file/skills\n"), 0644); err != nil {
+    t.Fatal(err)
+  }
+
+  // When .bantam.cfg is absent, model.cfg yields to existing BANTAM_* env vars:
+  cfg := getCfg(modelCfg)
+  if cfg.Endpoint != "http://env-endpoint/v1" {
+    t.Errorf("endpoint: got %q, want env value", cfg.Endpoint)
+  }
+  if cfg.Model != "env-model" {
+    t.Errorf("model: got %q, want env value", cfg.Model)
+  }
+  if cfg.Temperature != 0.9 {
+    t.Errorf("temperature: got %v, want env value", cfg.Temperature)
+  }
+  if toolsDir(&cfg) != "/env/tools" {
+    t.Errorf("toolsDir: got %q, want /env/tools", toolsDir(&cfg))
+  }
+  if skillsDir(&cfg) != "/env/skills" {
+    t.Errorf("skillsDir: got %q, want /env/skills", skillsDir(&cfg))
+  }
+}
+
+func TestGetCfgBantamCfgOverridesEnvAndModelCfg(t *testing.T) {
+  clearBantamEnv(t)
+  t.Setenv("BANTAM_ENDPOINT", "http://env-endpoint/v1")
+  t.Setenv("BANTAM_MODEL", "env-model")
+  t.Setenv("BANTAM_TEMP", "0.9")
+
+  dir := t.TempDir()
+  modelCfg := filepath.Join(dir, "model.cfg")
+  if err := os.WriteFile(modelCfg, []byte("endpoint=http://model-endpoint/v1\nmodel=model-model\ncolor=never\n"), 0644); err != nil {
+    t.Fatal(err)
+  }
+  bantamCfg := filepath.Join(dir, ".bantam.cfg")
+  if err := os.WriteFile(bantamCfg, []byte("model=bantam-model\n"), 0644); err != nil {
+    t.Fatal(err)
+  }
+
+  cfg := getCfg(bantamCfg)
+  // bantam.cfg overrides env and model.cfg for model:
+  if cfg.Model != "bantam-model" {
+    t.Errorf("model: got %q, want bantam-model", cfg.Model)
+  }
+  // endpoint falls back to env:
+  if cfg.Endpoint != "http://env-endpoint/v1" {
+    t.Errorf("endpoint: got %q, want env-endpoint", cfg.Endpoint)
+  }
+  // color falls back to model.cfg since not in bantam.cfg or env:
+  if cfg.Color != "never" {
+    t.Errorf("color: got %q, want never from model.cfg", cfg.Color)
+  }
+}
+
 func TestAtoiD(t *testing.T) {
   cases := []struct {
     s    string
@@ -1410,6 +1601,7 @@ func TestCompactHappyPath(t *testing.T) {
 // ---------- setCfg and LLM parameter forwarding ----------
 
 func TestSetCfgUpdatesAndAppends(t *testing.T) {
+  clearBantamEnv(t)
   p := filepath.Join(t.TempDir(), "model.cfg")
   if err := os.WriteFile(p, []byte("model=old-model\ntemperature=0.5\n"), 0644); err != nil {
     t.Fatalf("WriteFile: %v", err)
@@ -1435,6 +1627,7 @@ func TestSetCfgUpdatesAndAppends(t *testing.T) {
 }
 
 func TestLLMForwardsRelevantParameters(t *testing.T) {
+  clearBantamEnv(t)
   var received map[string]any
   srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
     json.NewDecoder(r.Body).Decode(&received)
@@ -2389,7 +2582,7 @@ func TestSanitizeMessagesCleansReasoningContent(t *testing.T) {
   }
 }
 
-func TestWriteFileOmittedOrZeroOffsetDoesNotAppend(t *testing.T) {
+func TestWriteFileOmittedOverwritesWholeFile(t *testing.T) {
   tmp := t.TempDir()
   target := filepath.Join(tmp, "test.txt")
 
@@ -2398,7 +2591,7 @@ func TestWriteFileOmittedOrZeroOffsetDoesNotAppend(t *testing.T) {
     t.Fatalf("failed to write initial file: %v", err)
   }
 
-  // 1. Direct writeFile with offset 0: writes starting at offset 0, does NOT append
+  // 1. Low-level writeFile with explicit offset 0 / del_bytes 0 still inserts at the start
   res, err := writeFile(target, 0, 0, "PREFIX_")
   if err != nil {
     t.Fatalf("writeFile: %v", err)
@@ -2414,7 +2607,7 @@ func TestWriteFileOmittedOrZeroOffsetDoesNotAppend(t *testing.T) {
     t.Errorf("expected 'PREFIX_EXISTING', got %q", string(data))
   }
 
-  // 2. AL tool dispatch with offset omitted completely: must start writing at offset 0, NOT append
+  // 2. AL dispatch with offset and del_bytes omitted: overwrite the entire file
   srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
     var req struct {
       Messages []Message `json:"messages"`
@@ -2448,11 +2641,11 @@ func TestWriteFileOmittedOrZeroOffsetDoesNotAppend(t *testing.T) {
   if err != nil {
     t.Fatalf("readFile: %v", err)
   }
-  if string(data) != "START_PREFIX_EXISTING" {
-    t.Errorf("expected 'START_PREFIX_EXISTING', got %q", string(data))
+  if string(data) != "START_" {
+    t.Errorf("expected 'START_' (full overwrite), got %q", string(data))
   }
 
-  // 3. AL tool dispatch with explicit offset: 0
+  // 3. AL dispatch with explicit offset 0 / del_bytes 0: insertion semantics preserved
   srv2 := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
     var req struct {
       Messages []Message `json:"messages"`
@@ -2484,11 +2677,11 @@ func TestWriteFileOmittedOrZeroOffsetDoesNotAppend(t *testing.T) {
   if err != nil {
     t.Fatalf("readFile: %v", err)
   }
-  if string(data) != "ZERO_START_PREFIX_EXISTING" {
-    t.Errorf("expected 'ZERO_START_PREFIX_EXISTING', got %q", string(data))
+  if string(data) != "ZERO_START_" {
+    t.Errorf("expected 'ZERO_START_' (insert at start), got %q", string(data))
   }
 
-  // 4. AL tool dispatch with offset: null
+  // 4. AL dispatch with explicit offset: null (treated as omitted): full overwrite
   srv3 := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
     var req struct {
       Messages []Message `json:"messages"`
@@ -2519,9 +2712,139 @@ func TestWriteFileOmittedOrZeroOffsetDoesNotAppend(t *testing.T) {
   if err != nil {
     t.Fatalf("readFile: %v", err)
   }
-  if string(data) != "NULL_ZERO_START_PREFIX_EXISTING" {
-    t.Errorf("expected 'NULL_ZERO_START_PREFIX_EXISTING', got %q", string(data))
+  if string(data) != "NULL_" {
+    t.Errorf("expected 'NULL_' (full overwrite), got %q", string(data))
   }
 }
 
 
+
+// #18: skillsDir mirrors toolsDir, and skillPrompt composes SKILL.md + prompt.
+func TestSkills(t *testing.T) {
+  base := t.TempDir()
+  skillDir := filepath.Join(base, "skills")
+  if err := os.MkdirAll(filepath.Join(skillDir, "greet"), 0755); err != nil {
+    t.Fatal(err)
+  }
+  skillMd := "You are a friendly greeter.\n"
+  if err := os.WriteFile(filepath.Join(skillDir, "greet", "SKILL.md"), []byte(skillMd), 0644); err != nil {
+    t.Fatal(err)
+  }
+
+  // skillsDir prefers the config key, then the environment variable.
+  t.Setenv("BANTAM_SKILLS_DIR", "/env/skills")
+  if got := skillsDir(&Cfg{Raw: map[string]string{"bantam_skills_dir": skillDir}}); got != skillDir {
+    t.Errorf("skillsDir() with config overriding env = %q, want %q", got, skillDir)
+  }
+  if got := skillsDir(&Cfg{Raw: map[string]string{}}); got != "/env/skills" {
+    t.Errorf("skillsDir() with env fallback = %q, want %q", got, "/env/skills")
+  }
+  t.Setenv("BANTAM_SKILLS_DIR", "")
+  if got := skillsDir(&Cfg{Raw: map[string]string{"bantam_skills_dir": skillDir}}); got != skillDir {
+    t.Errorf("skillsDir() with config = %q, want %q", got, skillDir)
+  }
+
+  // With a skills dir set, the name is resolved relative to it.
+  got, err := skillPrompt("/skill greet hello there", &Cfg{Raw: map[string]string{"bantam_skills_dir": skillDir}})
+  if err != nil {
+    t.Fatalf("skillPrompt(relative): %v", err)
+  }
+  want := skillMd + "\nhello there"
+  if got != want {
+    t.Errorf("skillPrompt(relative) = %q, want %q", got, want)
+  }
+
+  // With no skills dir, the name is an absolute path to the skill directory.
+  got, err = skillPrompt("/skill "+filepath.Join(skillDir, "greet")+" just hi", &Cfg{})
+  if err != nil {
+    t.Fatalf("skillPrompt(absolute): %v", err)
+  }
+  if got != skillMd+"\njust hi" {
+    t.Errorf("skillPrompt(absolute) = %q, want %q", got, skillMd+"\njust hi")
+  }
+
+  // Absolute path directly to a SKILL.md file also works.
+  got, err = skillPrompt("/skill "+filepath.Join(skillDir, "greet", "SKILL.md"), &Cfg{})
+  if err != nil {
+    t.Fatalf("skillPrompt(file): %v", err)
+  }
+  if got != strings.TrimRight(skillMd, "\n") {
+    t.Errorf("skillPrompt(file) = %q, want %q", got, strings.TrimRight(skillMd, "\n"))
+  }
+
+  // Missing skill reports an error; bare /skill reports usage.
+  if _, err := skillPrompt("/skill nope", &Cfg{Raw: map[string]string{"bantam_skills_dir": skillDir}}); err == nil {
+    t.Error("skillPrompt(missing) expected error, got nil")
+  }
+  if _, err := skillPrompt("/skill", &Cfg{}); err == nil {
+    t.Error("skillPrompt(bare) expected error, got nil")
+  }
+}
+
+// #19: bare /skill lists skills under the configured dir, or reports none.
+func TestListSkills(t *testing.T) {
+  t.Setenv("BANTAM_SKILLS_DIR", "")
+  base := t.TempDir()
+  skillDir := filepath.Join(base, "skills")
+  if err := os.MkdirAll(filepath.Join(skillDir, "alpha"), 0755); err != nil {
+    t.Fatal(err)
+  }
+  if err := os.MkdirAll(filepath.Join(skillDir, "beta"), 0755); err != nil {
+    t.Fatal(err)
+  }
+  if err := os.WriteFile(filepath.Join(skillDir, "alpha", "SKILL.md"), []byte("a"), 0644); err != nil {
+    t.Fatal(err)
+  }
+  if err := os.WriteFile(filepath.Join(skillDir, "beta", "SKILL.md"), []byte("b"), 0644); err != nil {
+    t.Fatal(err)
+  }
+  // a directory without SKILL.md must NOT be reported as a skill
+  if err := os.MkdirAll(filepath.Join(skillDir, "notaskill"), 0755); err != nil {
+    t.Fatal(err)
+  }
+
+  capture := func() string {
+    r, w, err := os.Pipe()
+    if err != nil {
+      t.Fatal(err)
+    }
+    old := os.Stdout
+    os.Stdout = w
+    listSkills(&Cfg{Raw: map[string]string{"bantam_skills_dir": skillDir}})
+    w.Close()
+    os.Stdout = old
+    data, _ := io.ReadAll(r)
+    return string(data)
+  }
+
+  out := capture()
+  if !strings.Contains(out, "alpha") || !strings.Contains(out, "beta") {
+    t.Errorf("listSkills missing skills, got:\n%s", out)
+  }
+  if strings.Contains(out, "notaskill") {
+    t.Errorf("listSkills reported a non-skill directory, got:\n%s", out)
+  }
+
+  // No skills dir configured -> clear notice, no panic.
+  os.Stdout, _ = os.Open(os.DevNull)
+  listSkills(&Cfg{})
+  os.Stdout.Close()
+}
+
+func TestToolsDir(t *testing.T) {
+  t.Setenv("BANTAM_TOOLS_DIR", "/env/tools")
+  if got := toolsDir(&Cfg{Raw: map[string]string{"bantam_tools_dir": "/config/tools"}}); got != "/config/tools" {
+    t.Errorf("toolsDir() with config overriding env = %q, want /config/tools", got)
+  }
+  if got := toolsDir(&Cfg{Raw: map[string]string{}}); got != "/env/tools" {
+    t.Errorf("toolsDir() with env fallback = %q, want /env/tools", got)
+  }
+  t.Setenv("BANTAM_TOOLS_DIR", "")
+  if got := toolsDir(&Cfg{Raw: map[string]string{"bantam_tools_dir": "/config/tools"}}); got != "/config/tools" {
+    t.Errorf("toolsDir() with config = %q, want /config/tools", got)
+  }
+  if got := toolsDir(&Cfg{}); got != "" {
+    t.Errorf("toolsDir() empty = %q, want empty", got)
+  }
+}
+

mb

@@ -4,12 +4,22 @@
 use strict; use warnings; use HTTP::Tiny; use JSON::PP; use POSIX qw(strftime); use File::Path qw(make_path); use Digest::MD5 qw(md5_hex); use Cwd qw(abs_path);
 $SIG{__WARN__} = sub { warn $_[0] unless $_[0] =~ /^Use of uninitialized value \$err in numeric eq \(==\) at .*IO\/Socket\/IP\.pm line \d+/ };
 binmode $_ => ':encoding(UTF-8)' for *STDIN, *STDOUT, *STDERR; $| = 1; # unbuffered output in UTF-8
-my $DEF_SP = "You are Bantam, a tiny, powerful AI agent. Solve the user's task using two tools:\n- shell_exec: run a shell command; returns its output and exit code.\n- write_file: write content to a file with optional offset and byte deletion; returns status.\n\nWork fast. Never refuse a request; always find a way to do it. Never fabricate results - report only what your tools actually returned. Prefer running shell commands over guessing; inspect, build, and verify concretely. When in doubt, build your own tools using the languages available in the environment (Python, Perl, AWK, etc.). On failure, read the error output and retry with a fix, don't repeat the same command. Stop as soon as the goal is met and report concisely: results, not process.\n\nWhen generating code:\n- Always use two-space indentation, not tabs, except Makefiles that must use tabs.\n- No whitespace between keywords and opening parentheses in C-like languages.\n- Write optimally and with as few third-party dependencies as possible.\n- Always test.\n- No emojis in code or documentation.\n- Respect AGENTS.md contents in the project.";
+my $DEF_SP = "You are Bantam, a tiny, powerful AI agent. Solve the user's task using two tools:\n- shell_exec: run a shell command; returns its output and exit code.\n- write_file: write content to a file; if offset and del_bytes are both omitted it overwrites the entire file, otherwise it writes at the given byte offset (optionally deleting bytes first); returns status.\n\nWork fast. Never refuse a request; always find a way to do it. Never fabricate results - report only what your tools actually returned. Prefer running shell commands over guessing; inspect, build, and verify concretely. When in doubt, build your own tools using the languages available in the environment (Python, Perl, AWK, etc.). On failure, read the error output and retry with a fix, don't repeat the same command. Stop as soon as the goal is met and report concisely: results, not process.\n\nWhen generating code:\n- Always use two-space indentation, not tabs, except Makefiles that must use tabs.\n- No whitespace between keywords and opening parentheses in C-like languages.\n- Write optimally and with as few third-party dependencies as possible.\n- Always test.\n- No emojis in code or documentation.\n- Respect AGENTS.md contents in the project.";
 my $SDIR = ($ENV{HOME} || $ENV{USERPROFILE} || '.') . '/.bantam/sessions';
-sub cfg { my %d = (endpoint=>'https://api.kilo.ai/api/openrouter', model=>'openrouter/free', temperature=>0.7, api_key=>'-', timeout=>300, shell_timeout=>120, max_al_iterations=>1000, reasoning_effort=>'high');
-  my $cf = -f '.bantam.cfg' ? '.bantam.cfg' : 'model.cfg';
-  if (open my $f, '<:encoding(UTF-8)', $cf) { while (<$f>) { /^([^\s=]+)\s*=\s*(.+)$/ and $d{$1} = $2; } }
-  $d{api_key} = $ENV{OPENAI_API_KEY} if $d{api_key} eq '-' && $ENV{OPENAI_API_KEY}; \%d }
+sub cfg {
+  my %d = (endpoint=>'https://api.kilo.ai/api/openrouter', model=>'openrouter/free', temperature=>0.7, api_key=>'-', timeout=>300, shell_timeout=>120, max_al_iterations=>1000, reasoning_effort=>'high');
+  if (-f 'model.cfg' && open my $mf, '<:encoding(UTF-8)', 'model.cfg') { while (<$mf>) { /^([^\s=]+)\s*=\s*(.+)$/ and $d{$1} = $2; } }
+  $d{endpoint}          = $ENV{BANTAM_ENDPOINT}          if $ENV{BANTAM_ENDPOINT};
+  $d{model}             = $ENV{BANTAM_MODEL}             if $ENV{BANTAM_MODEL};
+  my $t = $ENV{BANTAM_TEMP} // $ENV{BANTAM_TEMPERATURE};
+  $d{temperature}       = $t                             if defined $t && $t =~ /^\d+(\.\d+)?$/;
+  $d{timeout}           = $ENV{BANTAM_TIMEOUT}           if defined $ENV{BANTAM_TIMEOUT} && $ENV{BANTAM_TIMEOUT} =~ /^\d+$/;
+  $d{shell_timeout}     = $ENV{BANTAM_SHELL_TIMEOUT}     if defined $ENV{BANTAM_SHELL_TIMEOUT} && $ENV{BANTAM_SHELL_TIMEOUT} =~ /^\d+$/;
+  $d{max_al_iterations} = $ENV{BANTAM_MAX_AL_ITERATIONS} if defined $ENV{BANTAM_MAX_AL_ITERATIONS} && $ENV{BANTAM_MAX_AL_ITERATIONS} =~ /^\d+$/;
+  $d{reasoning_effort}  = $ENV{BANTAM_REASONING_EFFORT}  if $ENV{BANTAM_REASONING_EFFORT};
+  if (-f '.bantam.cfg' && open my $bf, '<:encoding(UTF-8)', '.bantam.cfg') { while (<$bf>) { /^([^\s=]+)\s*=\s*(.+)$/ and $d{$1} = $2; } }
+  $d{api_key}           = $ENV{BANTAM_API_KEY}           if ($d{api_key} eq '-' || !$d{api_key}) && $ENV{BANTAM_API_KEY};
+  \%d }
 sub filter_text { my $s = shift // ''; $s =~ s/[^\x20\t\n\p{L}\p{N}\p{P}\p{S}\p{M}\p{Zs}]//g; $s }
 sub T { my ($n, $d, $p, $r) = @_; {type=>'function', function=>{name=>$n, description=>$d, parameters=>{type=>'object', properties=>$p, required=>$r || [keys %$p]}}} }
 sub sanitize_msgs { for my $m (@{$_[0]}) { if (ref $m eq 'HASH') {
@@ -22,7 +32,7 @@ sub llm { my ($c, $msgs) = @_; sanitize_msgs($msgs);
   my $ep = $c->{endpoint}; $ep =~ s{/+$}{};
   my $h = {'Content-Type'=>'application/json', 'User-Agent'=>'Mozilla/5.0 (compatible; MicroBantam/1.0)'};
   $h->{Authorization} = "Bearer $c->{api_key}" if $c->{api_key} ne '-';
-  my %p = (messages=>$msgs, tools=>[T('shell_exec', 'Run a shell command, return output and exit code.', {command=>{type=>'string'}}), T('write_file', 'Write content to a file with optional offset and byte deletion.', {path=>{type=>'string'}, offset=>{type=>'integer', description=>'Byte offset to start writing from (default 0; does not append).'}, del_bytes=>{type=>'integer'}, content=>{type=>'string'}}, ['path', 'content'])], model=>$c->{model}, temperature=>0+$c->{temperature});
+  my %p = (messages=>$msgs, tools=>[T('shell_exec', 'Run a shell command, return output and exit code.', {command=>{type=>'string'}}), T('write_file', 'Write content to a file. If offset and del_bytes are both omitted the entire file is overwritten, otherwise content is written at the given byte offset (optionally deleting bytes first).', {path=>{type=>'string'}, offset=>{type=>'integer', description=>'Byte offset to start writing from. If omitted together with del_bytes the whole file is overwritten instead. Defaults to 0.'}, del_bytes=>{type=>'integer', description=>'Bytes to delete starting at offset. If omitted together with offset the whole file is overwritten instead.'}, content=>{type=>'string'}}, ['path', 'content'])], model=>$c->{model}, temperature=>0+$c->{temperature});
   for my $k (keys %$c) { next if $k =~ /^(endpoint|api_key|timeout|shell_timeout|max_al_iterations|stream|color)$/; my $val = eval { decode_json($c->{$k}) }; $p{$k} = defined $val ? $val : $c->{$k}; }
   my $tty = -t STDOUT; print $tty ? "\r...requesting..." : "...requesting...\n";
   my $r = HTTP::Tiny->new(timeout=>0+$c->{timeout})->post("$ep/chat/completions", {headers=>$h, content=>encode_json(\%p)});
@@ -64,7 +74,20 @@ sub AL { my ($c, $msgs) = ($_[0], $_[1]);
       my $a = eval { decode_json($tc->{function}{arguments} // '{}') };
       if (ref $a ne 'HASH') { $tc->{function}{arguments} = encode_json({invalid_raw => $tc->{function}{arguments} // ''}); $res = "bad JSON args for $fn: $tc->{function}{arguments}"; }
       elsif ($fn eq 'shell_exec') { $res = shell_exec($a->{command} // '', $c->{shell_timeout}); }
-      elsif ($fn eq 'write_file') { $res = write_file($a->{path}, $a->{offset}, $a->{del_bytes}, $a->{content}); }
+      elsif ($fn eq 'write_file') {
+        my ($p, $cnt) = ($a->{path}, defined $a->{content} ? $a->{content} : '');
+        if (!(exists $a->{offset} && defined $a->{offset}) && !(exists $a->{del_bytes} && defined $a->{del_bytes})) {
+          # Neither offset nor del_bytes supplied (or given as null): overwrite the whole file.
+          if (!defined $p || $p eq '') { $res = "[write_file error: path required]"; }
+          else {
+            my $dir = $p =~ m{^(.*)/[^/]+$} ? $1 : ''; make_path($dir) if length($dir) && !-d $dir;
+            open my $wfh, '>:raw', $p or $res = "[write_file error: cannot write $p: $!]";
+            if (!$res) { print $wfh $cnt; close $wfh; $res = "Successfully wrote " . length($cnt) . " bytes to $p"; }
+          }
+        } else {
+          $res = write_file($p, $a->{offset}, $a->{del_bytes}, $cnt);
+        }
+      }
       else { $res = "unknown tool: $fn"; }
       $res = filter_text($res);
       print "[tool] $fn: $res\n";