Mgeeeek
8cb4f09ded
Six independent fixes that make hook behaviour deterministic, debuggable, and free of silent footguns. Persist session_id in stored memories on_pre_compact.py now extracts session_id from the hook input and writes it under metadata.session_id, enabling search-by-session. Skip SessionStart bootstrap when MEM0_API_KEY is missing on_session_start.sh exits 0 early instead of emitting bootstrap text that instructs the agent to call mem0 MCP tools that would all fail without a key. Tighten block_memory_write.sh regex The old `*/memory/*.md` pattern blocked legitimate paths like docs/memory/architecture.md. New pattern blocks `*/MEMORY.md` and `*/.claude/memory/*` only -- the actual surfaces we want to protect. Opt-in debug logging via MEM0_DEBUG=1 Each hook (bash + python) redirects stderr / adds a FileHandler to ~/.mem0/hooks.log when MEM0_DEBUG is set. No helper module, no rotation -- KISS. Default behaviour is unchanged. Drop the duplicate PreCompact registration hooks.json and cursor-hooks.json had on_pre_compact.sh AND on_pre_compact.py registered on PreCompact, producing duplicate session-state writes. Drop the python entry. The python script stays alive -- on_stop.sh still spawns it as a session-end safety net. CI nudge Adds a single comment line to pyproject.toml because the repo's ci.yml path filters exclude mem0-plugin/-only PRs but the build_mem0/build_embedchain status checks are required by branch protection. Track the proper fix (path-conditional CI) as a follow-up.
46 lines
1.8 KiB
Bash
Executable File
46 lines
1.8 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# Hook: Stop
|
|
#
|
|
# Fires when Claude finishes responding.
|
|
# Reminds Claude to store any unsaved learnings, then spawns a background
|
|
# process to capture transcript state via the Mem0 REST API directly.
|
|
#
|
|
# Input: JSON on stdin with stop_hook_active, transcript_path, cwd
|
|
# Output: Text that becomes Claude's context (exit 0), or nothing
|
|
#
|
|
# IMPORTANT: Check stop_hook_active to avoid infinite loops.
|
|
|
|
set -euo pipefail
|
|
|
|
if [ -n "${MEM0_DEBUG:-}" ]; then
|
|
mkdir -p "$HOME/.mem0" && exec 2>>"$HOME/.mem0/hooks.log"
|
|
fi
|
|
|
|
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
|
|
|
|
INPUT=$(cat)
|
|
STOP_HOOK_ACTIVE=$(echo "$INPUT" | jq -r '.stop_hook_active // false' 2>/dev/null || echo "false")
|
|
|
|
if [ "$STOP_HOOK_ACTIVE" = "true" ]; then
|
|
exit 0
|
|
fi
|
|
|
|
cat <<'EOF'
|
|
Before finishing, check if there are important learnings from this interaction that should be persisted using the mem0 `add_memory` tool:
|
|
|
|
1. Were any significant decisions made? -> Store with metadata `{"type": "decision"}`
|
|
2. Were any new patterns or strategies discovered? -> Store with metadata `{"type": "task_learning"}`
|
|
3. Did any approach fail? -> Store with metadata `{"type": "anti_pattern"}`
|
|
4. Did you learn anything about the user's preferences? -> Store with metadata `{"type": "user_preference"}`
|
|
5. Were there environment/setup discoveries? -> Store with metadata `{"type": "environmental"}`
|
|
|
|
Memories can be as detailed as needed — include full context, reasoning, code snippets, file paths, and examples. Longer, searchable memories are more valuable than vague one-liners.
|
|
|
|
If nothing notable happened in this interaction, it's fine to skip. Only store genuinely useful learnings.
|
|
EOF
|
|
|
|
# Capture transcript state in the background via Mem0 REST API
|
|
echo "$INPUT" | python3 "$SCRIPT_DIR/on_pre_compact.py" --source=session-end 2>/dev/null &
|
|
|
|
exit 0
|