agentleFS
shubham303shubham303/meelu-analytics-mcp/.claude/skills/fix-bug/SKILL.md

fix-bug

Skill9 starsChanged 38 days agoMITView on GitHub

Autonomously pick a random tool of the meelu-analytics MCP server, test it end-to-end through the protocol and with a real agent on realistic and adversarial data, and if anything is broken, find the root cause, fix it, and open a PR. Use when asked to "hunt bugs", "test a random feature", or for the daily autonomous bug run. Optional argument - a tool name to test instead of a random one.

  • Commits and pushes

45 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use improvement, build-feature and 10 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.
  • kiminmonacoSkill

    improvement

    Guide for making improvements to existing ClaudeBar functionality using TDD. Use this skill when: (1) Enhancing existing features (not adding new ones) (2) Improving UX, performance, or code quality (3) User asks "improve X", "make Y better", or "enhance Z" (4) Small enhancements that don't require full architecture design For NEW features, use implement-feature skill instead.

    Also in 4 projects that use this

  • sirakinbSkill

    build-feature

    Autonomous task loop that picks ready tasks, implements them, updates progress.txt, commits, and repeats. Use when asked to 'build feature', 'run the loop', or 'implement these tasks'.

    Also in 5 projects that use this

  • kiminmonacoSkill

    implement-feature

    Guide for implementing features in ClaudeBar following architecture-first design, TDD, rich domain models, and Swift 6.2 patterns. Use this skill when: (1) Adding new functionality to the app (2) Creating domain models that follow user's mental model (3) Building SwiftUI views that consume domain models directly (4) User asks "how do I implement X" or "add feature Y" (5) Implementing any feature that spans Domain, Infrastructure, and App layers

    Also in 7 projects that use this

  • SentrySkill

    write-tests

    Write tests following project conventions. Use when adding new tests or modifying existing ones.

    Also in 6 projects that use this

  • JetBrainsSkill

    create-plan

    Research the codebase and create an implementation plan with architecture notes, design document, and track decomposition. Use when starting a new feature or large change.

    Also in 4 projects that use this

  • JasonPaffSkill

    implement-plan

    Execute implementation plan with structured tracking and validation using subagent architecture

    Also in 3 projects that use this

  • Other versions45 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 45

    1. remix-run/react-router★ 57ksame name
    2. ktorio/ktor★ 15ksame name
    3. JetBrains/Exposed★ 9.3ksame name
    4. bytebase/dbhub★ 3.9ksame name
    5. The-OpenROAD-Project/OpenROAD★ 3.3ksame name
    6. JamieMason/syncpack★ 2.1ksame name
    7. tddworks/ClaudeBar★ 1.6ksame name
    8. Harry-kp/vortix★ 757same name
    9. Doist/todoist-ai★ 563same name
    10. tddworks/asc-cli★ 393same name

    35 more not listed

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repositoryOne other file it gives its agents.

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 45 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the working tree is dirty, stop and report.

    line 23
  2. Very specific:

    If $ARGUMENTS names a tool, use it.

    line 33
  3. Very specific:

    Run them through the real protocol with uv run python scripts/e2e/mcp_call.py --steps <file> (prerequisite calls like create_session, join, …

    line 50
  4. Distinctive:

    If you find nothing after a genuine effort (at least 6 distinct scenarios + the agent run), test one more random tool.

    line 67
  5. Distinctive:

    Keep the fix minimal and in the surrounding style.

    line 79
  6. Very specific:

    If you found several bugs, fix the most severe one in this PR and list the rest in the PR body (and add them to docs/BACKLOG.md as pending, …

    line 84
  7. Very specific:

    Commit fix(<area>): <what was wrong> ending with: Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

    line 95

“Common” is measured: kept by at least 6 in 10 of the 45 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

Test one random feature end to end; fix what's broken

You are the daily bug-hunting agent for meelu-analytics-mcp. Pick one tool, try hard to break it the way real users and agents would, and if it's broken, fix it properly. Never ask the user questions. Never merge, never push to main, never force-push.

0. Preflight

git fetch origin && git checkout main && git pull --ff-only
gh label create autonomous --color 5319e7 --description "Opened by an autonomous agent" 2>/dev/null || true
gh label create bug --color d73a4a 2>/dev/null || true
uv sync --extra dev --extra insights
uv run pytest -q

If the working tree is dirty, stop and report. If pytest already fails on main, that failure is today's bug — go straight to step 3 with it.

1. Pick the target

uv run python scripts/e2e/mcp_call.py --list > /tmp/meelu-tools.txt
gh pr list --state open --label autonomous --json title,headRefName

If $ARGUMENTS names a tool, use it. Otherwise choose randomly (python3 -c 'import random,sys; print(random.choice(open(sys.argv[1]).read().split("\n")[:-1]).split(":")[0])' /tmp/meelu-tools.txt), re-rolling if the tool already has an open autonomous PR. Read the tool's function in src/tabint/analysis/tools.py, the algorithm behind it, and its page under docs/tools/, so you know what it promises.

2. Try to break it

Generate datasets with pandas/numpy in a temp dir. Always include one realistic dataset where you know the right answer (planted effect, known clusters, known trend), plus several adversarial ones chosen for this tool, e.g.: nulls in key columns, a constant column, a single row, two rows, all-unique IDs, high-cardinality categoricals, unicode/space-laden column names, numeric strings, mixed date formats, timezone-aware timestamps, duplicate column names, an empty CSV with only a header, very skewed or tiny classes, huge values, negative values where only positives make sense.

Run them through the real protocol with uv run python scripts/e2e/mcp_call.py --steps <file> (prerequisite calls like create_session, join, set_column_type go first; "$session_key" is substituted). Then ask a real agent a natural question that needs the tool: scripts/e2e/agent_e2e.sh "<question with absolute CSV path>" (if the claude CLI is unavailable, note it and rely on the protocol runs).

A run is a bug if any of these hold:

  • a tool error/traceback where a clear declined result was due;
  • a wrong number versus your independent pandas/scipy/sklearn computation;
  • non-determinism (same input twice → different method or output);
  • the result's metadata doesn't record the chosen method, or trust is missing, overconfident (e.g. high on 3 rows), or declined without a reason;
  • the tool description/docs promise something the tool doesn't do, such that the real agent misuses it or gives a wrong answer;
  • a file-access escape outside TABULAR_BASE / the data root.

If you find nothing after a genuine effort (at least 6 distinct scenarios + the agent run), test one more random tool. If both are clean, stop: report what you tested and the scenarios, and open no PR.

3. Fix it

  1. git checkout -b auto/fix-<tool>-<short-slug>
  2. Reproduce first: add a pytest in tests/ that fails on the current code (follow tests/test_containment.py for fixtures and the _tool(name) helper).
  3. Find the root cause — don't patch the symptom at the tool boundary if the bug lives in the algorithm or a shared helper. Check whether sibling tools share the faulty code path and fix them too (with tests).
  4. Keep the fix minimal and in the surrounding style. Keep both invariants: deterministic method selection recorded in metadata; honest trust/declined.
  5. Re-run: uv run pytest -q, the failing protocol steps, and the agent question. All must now pass.

If you found several bugs, fix the most severe one in this PR and list the rest in the PR body (and add them to docs/BACKLOG.md as pending, Type: bug).

4. Independent review

Launch a general-purpose subagent with the bug description, the reproduction, and git diff main...HEAD; ask it to check the root-cause claim, look for regressions and missed sibling code paths. Fix real findings and re-verify.

5. Ship

  • Commit fix(<area>): <what was wrong> ending with: Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
  • git push -u origin HEAD
  • gh pr create --base main --label autonomous --label bug --title "fix(<tool>): ..." --body ... Body: Tool tested, Scenarios tried (table: scenario → result), Bug (exact input, expected vs actual), Root cause, Fix, Verification (new test, pytest summary, protocol excerpt, agent question
    • answer before/after), Other issues found. End the body with: 🤖 Generated with [Claude Code](https://claude.com/claude-code)

6. Report

Finish with: tool(s) tested, scenarios, bug found (or "none"), PR URL, and anything a human should double-check.

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.