#!/usr/bin/env bash

# scripts/post_commit_remind.sh

#

# PostToolUse hook for Bash. When `git commit` just ran, re-inject the

# autonomous-loop continuation reminder from CLAUDE.md § The only stop

# + § Autonomous Workflow "Loop: Step 0 → 7 → next task's Step 0".

#

# Why this exists: CLAUDE.md is loaded once per session; long sessions

# accumulate momentum drift where the loop's "only stop is user

# explicit stop" rule fades from active recall. Re-injecting the rule

# at every commit lands it at the exact moment the drift is most

# likely (just after Step 6, before the next task's Step 0).

#

# Inputs: stdin JSON from Claude Code (PostToolUse hook payload).

# stdout: reminder text fed back to Claude as tool result feedback.

set -u

INPUT=$(cat)

CMD=$(printf '%s' "$INPUT" | python3 -c '

import sys, json

try:

d = json.load(sys.stdin)

print(d.get("tool_input", {}).get("command", ""))

except Exception:

pass

' 2>/dev/null || true)

case "$CMD" in

*"git commit"*)

;;

*)

exit 0

;;

esac

REMINDER=$(cat <<'EOF'

[post-commit reminder]

A commit landed. Recall the project's fully-autonomous principle:

return to the head of the autonomous loop in CLAUDE.md

§ Autonomous Workflow and continue the next unit of work

immediately. Intermediate artefacts — per-task notes, status

summaries, planning text — are fine to produce on the way, but

must never become a reason to stop. Smell triggers are in-flight

interrupts; build / test failures are Active-task items; neither

halts the loop. Only the user's explicit stop directive halts

the loop (CLAUDE.md § The only stop).

EOF

)

# PostToolUse: stdout must be JSON with hookSpecificOutput.additionalContext

# for the text to be injected into Claude's next-turn context. Bare-text

# stdout is shown only in transcript mode (Ctrl-R), not to Claude.

export REMINDER

python3 -c '

import json, os

print(json.dumps({

"hookSpecificOutput": {

"hookEventName": "PostToolUse",

"additionalContext": os.environ.get("REMINDER", ""),

}

}))

'