Agent Log
Mantiene un diario de decisiones del repositorio que los agentes deben leer antes de cambiar el código cubierto
Probado · Funciona
Qué hace
Maintains .agents/log, un version-controlled journal de product y engineering decisions co-owned by the developer y el agent, con a fixed entry format, a wip/done lifecycle y an index table. Before non-trivial changes it makes the agent grep prior entries by related_paths y keywords, treat finalized entries as binding, y surface conflicts instead of silently reversing them. Se activa on explicit log requests y on feature work inside a repo that already has .agents/log; it never initializes the log on its own.
Informe de la prueba
Se construyó un repo cuyo log already contained a finalized entry choosing Redis para el job queue, then asked for a write-up del opposite decision. Without the skill el agent produced a readable page that never mentioned the decision it was reversing y left index.md stale, so the new entry fue effectively invisible; los two prescribed grep commands de la skill surfaced the contradicting entry immediately y el run linked back to it, kept the frozen entry byte-identical, y opened the new one as wip rather than done. A 22-point conformance check scored the plain write-up 4/22 y la skill version 22/22. Note el upstream rename: la skill formerly published as design-log es now agent-log, y el old path no longer exists in the repo.
Probado el: 2026-08-10 · Claude Code 2.x (agent harness)
Instalación
git clone https://github.com/denysdovhan/agents.git mkdir -p ~/.claude/skills cd agents && cp -r skills/agent-log ~/.claude/skills/agent-log
Comandos y prompts de ejemplo
/agent-logMantiene un diario de decisiones del repositorio que los agentes deben leer antes de cambiar el código cubierto
Los skills se activan con peticiones en lenguaje natural, sin comandos que memorizar. Tras instalarlo, prompts como estos lo activan (en inglés):
Log our decision to move background jobs to Postgres pg-bossWrite a .agents/log entry reversing the Redis job queue decisionCheck prior agent-log entries for context before this refactor