agent-self-reflection
от brennerspear
Web & Frontend DevelopmentКратко
Periodic self-reflection on recent sessions.
Описание
Что делает этот навык
Runs hourly as a cron job to analyze recent agent sessions, identifying what succeeded and what failed. Extracts specific, actionable lessons and routes each insight to the appropriate workspace file based on type. Skips vague summaries and only writes insights that are new, specific, and directly applicable.
It routes each insight to a specific file by type rather than appending everything to one log, so future sessions find the right context without sifting through noise.
Когда его использовать
- Catching repeated tool failures across multiple sessions
- Logging user preferences discovered during 1:1 conversations
- Updating daily memory files with project context from today's work
- Fixing skill instructions after a workflow breaks in practice
- Reviewing how automated cron tasks performed overnight
Пример сценария
User asks: Run self-reflection on the last 2 hours
- 1
Lists sessions active in the last 120 minutes, filtering out subagent sessions
- 2
Reads the last 50 lines of each session JSONL to stay within token budget
- 3
Analyzes each session for successes, errors, retries, and misunderstandings
- 4
Routes each insight to the correct file: AGENTS.md, TOOLS.md, memory/YYYY-MM-DD.md, or a skill file
- 5
Writes a 2-4 sentence summary of what was reviewed and what was recorded
Brief summary confirming which sessions were reflected on and which files were updated with new insights
Установка
clawhub install brennerspear/agent-self-reflectionВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.