add non-interactive mode and the subagent delegation prompt
Port the two latest changes of the Perl original. Non-interactive mode: a prompt as an argument runs that one task and exits. Every word is joined into a single task, so quoting is optional, and an empty or whitespace-only argument still opens the REPL. The ready banner moves to stderr so stdout carries the answer, progress and token stats alone, stdout is line buffered so a piped run arrives as it is written, and the process exits with the task status: the command's own for !cmd, 1 on failure, 130 on Ctrl-C. dispatch() returns an int for that. Subagent delegation: the system prompt now tells the model it can split a multi-stage task by copying .solo_sess.json into a fresh directory under the scratch dir and re-launching this binary from there with the same endpoint, model, tools and prompt, which the copy carries over. The prompt names our own path, resolved through /proc/self/exe, and notes that a subagent answers on stdout while its banner goes to stderr, so independent subtasks run in parallel with &. The prompt buffer is now sized from the expanded paths, and trailing slashes are stripped from SOLO_ENDPOINT as in the original, so the command in the prompt is always well formed. Also print the token stats only on the turn that ends the tool loop, as the original does: they used to land in the middle of a tool run and chop up whatever was piped out of a one-shot run. Tests go to 28, all offline: 15 new ones, including a loopback mock endpoint that drives a tool call and checks where the stats land.
This commit is contained in:
Luxferre
@@ -10,7 +10,9 @@
|
||||
set -u
|
||||
|
||||
SOLO=${1:-./solo}
|
||||
case $SOLO in /*) ;; *) SOLO=$PWD/$SOLO ;; esac
|
||||
# Canonical, like /proc/self/exe inside the agent, which names its own path to the
|
||||
# model so a subagent can be re-launched.
|
||||
SOLO=$(perl -MCwd -e 'print Cwd::abs_path($ARGV[0]) // $ARGV[0]' "$SOLO" 2>/dev/null) || SOLO=${1:-./solo}
|
||||
[ -x "$SOLO" ] || { echo "test: no such executable: $SOLO" >&2; exit 1; }
|
||||
command -v perl >/dev/null 2>&1 || { echo "test: perl is required" >&2; exit 1; }
|
||||
|
||||
@@ -40,6 +42,13 @@ solo_in() {
|
||||
( cd "$_dir" && env "$@" "$SOLO" )
|
||||
}
|
||||
|
||||
# solo_args DIR ARG... -- the one-shot mode: no stdin, no REPL
|
||||
solo_args() {
|
||||
_dir=$1; shift
|
||||
mkdir -p "$_dir"
|
||||
( cd "$_dir" && "$SOLO" "$@" < /dev/null )
|
||||
}
|
||||
|
||||
echo "# $SOLO"
|
||||
|
||||
# --- a setting taken from the environment must not be freed by the agent -------
|
||||
@@ -116,5 +125,97 @@ printf '/quit\n' | solo_in "$D" SOLO_TIMEOUT=42 >/dev/null 2>&1
|
||||
perl -0777 -ne 'exit(/after 42s is killed/ ? 0 : 1)' "$D/.solo_sess.json"
|
||||
ok $? 'the bash tool description quotes SOLO_TIMEOUT'
|
||||
|
||||
# --- a prompt as an argument runs one task and exits --------------------------
|
||||
# The banner moved to stderr, so stdout holds the answer alone and a one-shot run
|
||||
# can be piped into another program.
|
||||
D=$WORK/oneshot
|
||||
solo_args "$D" '!echo hello' >"$WORK/out" 2>"$WORK/err"
|
||||
ok $? 'a prompt argument runs the task and exits'
|
||||
grep -q '^hello$' "$WORK/out"
|
||||
ok $? 'the answer arrives on stdout'
|
||||
grep -q 'Agent Solo ready' "$WORK/err"
|
||||
ok $? 'the ready banner goes to stderr in one-shot mode'
|
||||
if grep -q 'Agent Solo ready' "$WORK/out"; then rc=1; else rc=0; fi
|
||||
ok "$rc" 'stdout carries no banner in one-shot mode'
|
||||
# Anything left on stdin is ignored: the one-shot run never opens the REPL.
|
||||
( cd "$D" && printf '!echo from-stdin\n/quit\n' | "$SOLO" '!echo hello' ) >>"$WORK/out" 2>>"$WORK/err"
|
||||
if grep -q 'from-stdin' "$WORK/out"; then rc=1; else rc=0; fi
|
||||
ok "$rc" 'a one-shot run never reads stdin'
|
||||
|
||||
solo_args "$WORK/onestatus" '!sh -c "exit 7"' >/dev/null 2>&1
|
||||
[ $? -eq 7 ]
|
||||
ok $? 'a ! task hands its own exit status back'
|
||||
|
||||
solo_args "$WORK/joined" '!echo' one two >"$WORK/joined.out" 2>/dev/null
|
||||
grep -q '^one two$' "$WORK/joined.out"
|
||||
ok $? 'every word becomes part of one task, quoting optional'
|
||||
|
||||
# An empty or whitespace-only argument is no task at all, so the REPL opens.
|
||||
mkdir -p "$WORK/blank"
|
||||
( cd "$WORK/blank" && printf '/quit\n' | "$SOLO" ' ' ) >"$WORK/blank.out" 2>/dev/null
|
||||
grep -q 'Agent Solo ready' "$WORK/blank.out"
|
||||
ok $? 'a whitespace-only argument still opens the REPL'
|
||||
|
||||
# --- the usage line belongs to the last turn, not to every turn ----------------
|
||||
# The token stats used to be printed after every turn, so they landed in the
|
||||
# middle of a tool run and chopped up whatever was piped out of a one-shot run.
|
||||
cat >"$WORK/mock.pl" <<'MOCK_EOF'
|
||||
# A chat endpoint on the loopback: the first turn asks for a tool, the second one
|
||||
# answers. Prints the port it got, then serves until it is killed.
|
||||
use IO::Socket::INET;
|
||||
my $srv = IO::Socket::INET->new(LocalAddr => '127.0.0.1', LocalPort => 0,
|
||||
Listen => 5, ReuseAddr => 1) or exit 1;
|
||||
$| = 1;
|
||||
print $srv->sockport, "\n";
|
||||
my $n = 0;
|
||||
while(my $c = $srv->accept) {
|
||||
$c->autoflush(1);
|
||||
my $req = '';
|
||||
while(my $l = <$c>) { last if $l =~ /^\r?\n$/; $req .= $l; }
|
||||
print $c "HTTP/1.0 100 Continue\r\n\r\n" if $req =~ /^Expect:/mi;
|
||||
my ($len) = $req =~ /Content-Length:\s*(\d+)/i;
|
||||
my $body = '';
|
||||
read($c, $body, $len) if $len;
|
||||
$n++;
|
||||
my $call = '{"role":"assistant","content":"","tool_calls":[{"id":"c1","type":"function",'
|
||||
. '"function":{"name":"bash","arguments":"{\"command\":\"echo tool-ran\"}"}}]}';
|
||||
my $msg = $n == 1 ? $call : '{"role":"assistant","content":"The answer is 42."}';
|
||||
my $out = '{"model":"mock/free","usage":{"total_tokens":7,"prompt_tokens":5,"completion_tokens":2},'
|
||||
. '"choices":[{"message":' . $msg . '}]}';
|
||||
print $c "HTTP/1.0 200 OK\r\nContent-Type: application/json\r\nContent-Length: "
|
||||
. length($out) . "\r\n\r\n" . $out;
|
||||
close $c;
|
||||
}
|
||||
MOCK_EOF
|
||||
perl "$WORK/mock.pl" >"$WORK/port" 2>/dev/null &
|
||||
MOCK=$!
|
||||
i=0
|
||||
while [ ! -s "$WORK/port" ] && [ "$i" -lt 50 ]; do sleep 0.1; i=$((i + 1)); done
|
||||
PORT=$(cat "$WORK/port" 2>/dev/null)
|
||||
mkdir -p "$WORK/tokens"
|
||||
( cd "$WORK/tokens" && SOLO_ENDPOINT="http://127.0.0.1:$PORT/api" "$SOLO" 'do it' ) \
|
||||
>"$WORK/tokens.out" 2>"$WORK/tokens.err"
|
||||
kill "$MOCK" 2>/dev/null
|
||||
[ "$(grep -c 'tokens:' "$WORK/tokens.out")" -eq 1 ]
|
||||
ok $? 'the token stats are printed once, on the last turn' \
|
||||
"$(grep -c 'tokens:' "$WORK/tokens.out") line(s)"
|
||||
awk '/tool-ran/{t=NR} /tokens:/{s=NR} END{exit !(t && s && s > t)}' "$WORK/tokens.out"
|
||||
ok $? 'the tool run is not chopped up by stats' "$(tail -1 "$WORK/tokens.out")"
|
||||
grep -q 'The answer is 42\.$' "$WORK/tokens.out"
|
||||
ok $? 'the answer of the last turn reaches stdout'
|
||||
tail -1 "$WORK/tokens.out" | grep -q '^--- \[mock/free\] tokens:'
|
||||
ok $? 'the stats line is the last thing on stdout'
|
||||
|
||||
# --- the model has to know how to delegate to a subagent ----------------------
|
||||
D=$WORK/subagent
|
||||
printf '/quit\n' | solo_in "$D" SOLO_ENDPOINT=http://e.example/api/// >/dev/null 2>&1
|
||||
perl -0777 -ne 'exit(/split into subagents/ ? 0 : 1)' "$D/.solo_sess.json"
|
||||
ok $? 'the system prompt tells the model it can spawn subagents'
|
||||
tr -d '\\' <"$D/.solo_sess.json" >"$D/plain" # json-c escapes every slash
|
||||
grep -qF "SOLO_ENDPOINT=http://e.example/api/ $SOLO 'the subtask, stated in full'" "$D/plain"
|
||||
ok $? 'the subagent line names the endpoint and Solo own path'
|
||||
if grep -q 'api///' "$D/.solo_sess.json"; then rc=1; else rc=0; fi
|
||||
ok "$rc" 'a trailing slash on SOLO_ENDPOINT is stripped'
|
||||
|
||||
printf '%d passed, %d failed\n' "$PASS" "$FAIL"
|
||||
[ "$FAIL" -eq 0 ]
|
||||
|
||||
Reference in New Issue
Block a user