agentleFS
naman10parikhnaman10parikh/pixel/.claude/skills/self-improve/SKILL.md

self-improve

Skill0 starsChanged 27 days agoMITView on GitHub

Analyze past session logs to find repeated patterns, forgotten rules, and backtracking moments. Auto-update CLAUDE.md, skills, and rules to address gaps. "Correct once, never again.

39 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use aeon-update, auto-workflow and 2 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • aaronjmarsSkill

    aeon-update

    Pull framework updates from the upstream Aeon repo into this instance - 3-way merges canon's new commits into a PR, never clobbering operator config.

    Also in 4 projects that use this

  • aaronjmarsSkill

    auto-workflow

    Two-mode aeon.yml workflow builder - analyze inspects URLs and emits a tiered, signal-verified skill-enablement plan plus an aeon.yml diff; enable flips slugs to enabled:true and opens a PR.

    Also in 4 projects that use this

  • +2 more from aaronjmars

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 2 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 29 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After several sessions of active development

    line 8
  2. Distinctive:

    When you notice Claude making the same mistakes repeatedly

    line 9
  3. Very specific:

    Read session files from ~/.claude/projects/ (JSONL format).

    line 16
  4. Very specific:

    Fix: Always console.warn with context prefix and error message

    line 69
  5. Very specific:

    RULE: Every catch block must either log or have // Intentionally silent: [reason]

    line 70

“Common” is measured: kept by at least 6 in 10 of the 29 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

When to Use

  • After several sessions of active development
  • When you notice Claude making the same mistakes repeatedly
  • Periodically (weekly) as a maintenance task

Process

1. Session Log Analysis

Read session files from ~/.claude/projects/ (JSONL format). Look for:

  • Questions Claude asks repeatedly (→ add answer to CLAUDE.md)
  • Rules Claude forgets (→ add to .claude/rules/)
  • Information Claude keeps re-discovering (→ promote to bootstrap memory)
  • Patterns of backtracking (→ create a skill that prevents it)
  • Tool call failures (→ improve error handling in hooks)

2. Signal Classification

  • High confidence: Explicit corrections from user → immediately persist
  • Medium confidence: Patterns that worked well → persist after confirming
  • Low confidence: Observations → queue for review

3. Output

For each finding:

  1. Identify which file to update (CLAUDE.md, a rule, a skill, settings)
  2. Make the minimal change needed
  3. Log the improvement to CONTEXT.md

4. Verification

  • Confirm CLAUDE.md is still under 80 lines / 2K tokens
  • Confirm no conflicting rules exist
  • Confirm skills are still discoverable (short descriptions)

Known Error Pattern Categories (Auto-Updated)

Category 1: Vercel Serverless State Loss

  • Pattern: In-memory Maps/globals lost between Lambda invocations
  • Files at risk: Any const map = new Map() at module scope in API routes
  • Fix: Client-side state (Zustand/localStorage) + reconnect via SDK on every request
  • Examples: ptyRegistry (terminal), desktopSessions (e2b-desktop)

Category 2: Race Conditions in E2B Sandbox Boot

  • Pattern: Operation attempted before dependency is ready (Chrome, agent, PTY)
  • Fix: Retry loop with backoff, never assume ready after fixed sleep
  • Examples: Chrome window streaming, agent-runner readiness, CDP port binding

Category 3: SSE/Stream Timeout During Long Operations

  • Pattern: Browser/proxy drops connection when no data flows for 10-15s
  • Fix: Keepalive pings on all SSE streams
  • Examples: Chat route during tool execution, terminal PTY during inactivity

Category 4: Silent Catch Blocks

  • Pattern: catch { } or catch { /* comment */ } with no logging
  • Fix: Always console.warn with context prefix and error message
  • RULE: Every catch block must either log or have // Intentionally silent: [reason]

Category 5: Missing Circuit Breakers

  • Pattern: Unlimited retries or cascading failures from flaky dependencies
  • Fix: Track failure count, open circuit after N failures, log transitions
  • Examples: Anthropic API rate limits, Supabase connectivity, E2B sandbox creation

The Compound Effect

Each improvement makes the next session slightly better. Over weeks, this compounds into a dramatically better development experience. This IS AutoLab applied to our own tooling.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.