Self-Review Skill

Free to use. If this saved you time, you can support development here: [PAYMENT_LINK]

Overview

This skill helps agents improve their own configuration by analyzing logged incidents and activity patterns, then proposing specific edits to enhance future performance. It examines AGENTS.md, SOUL.md, HEARTBEAT.md, MEMORY.md, TOOLS.md, USER.md, Manuel.md, and SKILLS.md.

Installation

Place this skill in your agent's skills directory (e.g., ./skills/self-review/)
The skill will automatically create a trainer/ directory at ../../trainer/ relative to itself on first run:
- If skill is at ./skills/self-review/, trainer will be created at ../../trainer/
- Adjust the path if your directory structure differs
The trainer directory contains all state needed for the self-review process:
- logs/incidents.md - incident log file
- config.js - configuration file (AUTO_APPLY: false/true)
- logs/.last-review - timestamp tracking last review run
- proposals/ - directory for review proposals awaiting approval
- backups/ - directory for file backups before applying changes

How It Works

When to run

Weekly, or when ../../trainer/logs/incidents.md has 10+ new entries since the last review
On manual request ("run self-review")

Step 1 — Gather

First-run setup: If ../../trainer/ doesn't exist, the skill automatically creates the complete folder structure:

Creates ../../trainer/logs/, ../../trainer/proposals/, ../../trainer/backups/ directories
Creates ../../trainer/config.md with: AUTO_APPLY: false
Creates ../../trainer/logs/.last-review with: 2000-01-01T00:00:00
Creates ../../trainer/logs/incidents.md with: 
Read every entry in ../../trainer/logs/incidents.md timestamped after the value in ../../trainer/logs/.last-review (create with far-past date if missing)
Skim HEARTBEAT.md and MEMORY.md for relevant activity not formally logged
If nothing new since last review, stop and report "nothing to review"

Step 2 — Classify

For each incident, note:

type: fail / friction / win
what happened, what was tried, what worked or didn't
best-guess target file:
- AGENTS.md — behavior rules, task workflow, decision logic
- SOUL.md — identity, values, tone, personality
- HEARTBEAT.md — recurring routines, check-in cadence
- MEMORY.md — durable facts worth remembering
- TOOLS.md — tool usage patterns, gotchas, correct invocations
- USER.md — user preferences, context, constraints
- Manuel.md — step-by-step how-tos / procedures
- SKILLS.md — skill index, when to use which skill

Step 3 — Draft proposed edits

Surgical, specific edits only — never a full-file rewrite
Every proposed edit must cite the incident(s) that justify it
Group by target file
Save as ../../trainer/proposals/<YYYY-MM-DD>-review.md, formatted like:

AGENTS.md

ADD under [section]: "..."
Reason: incidents #3, #7 — repeated X failure

(repeat per affected file)

Step 4 — Apply or hold

Read ../../trainer/config.md
If AUTO_APPLY: false (default) — stop here. Present the proposal to the agent operator in chat and wait for explicit approval before touching any live file.
If AUTO_APPLY: true — for each approved-category edit:
1. Copy the current file to ../../trainer/backups/<filename>.<timestamp>.bak
2. Apply the edit exactly as drafted — no improvising beyond what was proposed
3. Log what changed to ../../trainer/logs/applied.md
Regardless of AUTO_APPLY, SOUL.md and USER.md always require manual approval — never auto-edit identity or user-context files.

Step 5 — Close out

Update ../../trainer/logs/.last-review to the current timestamp
Summarize in plain language what was found and what changed (or is pending approval)

Configuration

The ../../trainer/config.md file controls behavior:

AUTO_APPLY: false

Set to true to enable automatic application of safe edits (excludes SOUL.md and USER.md which always require manual approval).

The system will automatically initialize required files with these defaults on first run:

../../trainer/logs/.last-review: 2000-01-01T00:00:00
../../trainer/logs/incidents.md:

Self Review Publish

安装技能