Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+22 more

    Skill Miner for AI Coding Agents

    1

    Turns repeated corrections in your Claude Code, Codex and Cursor transcripts into skills and rules you approve.

    Free

    Security scanned
    Skill Miner for AI Coding Agents

    Skill Miner for AI Coding Agents

    Example session with this skill installed

    Mine the sample transcripts from the last 30 days for skill candidates and show me what you find.

    • Read your context and instructions
    • Compiled the miner for ai

    Skill Miner on a synthetic sample. 7 sessions in 3 made-up projects (ledgerlite-api, pebble-web, harbor-etl), last 30 days. The counts below are the output of node scripts/mine.js --days 30; the proposals were drafted from that report.

    What it found

    1 repeated long instruction, typed in 3 sessions across 3 projects: "Prepare the next release: bump the version in the project file, move the Unreleased section of CHANGELOG.md under a new dated heading, commit it as chore(release), then tag it. Do not push."

    10 corrections. Eight of them fall into three groups: use TimeProvider, not DateTime.Now or UtcNow (3 sessions); use conventional commits (3 projects); use date-fns, not moment (2 sessions).

    5 shell recipes that ran in 3 sessions each: docker compose up -d db, dotnet ef database update, dotnet test ... --filter, pnpm lint --fix, pnpm test.

    2 tool sequences, led by Edit:CHANGELOG.md → Bash:git commit → Bash:git tag (3 sessions, 3 projects).

    Proposals for you to keep or drop

    NameTargetWhat it capturesSessionsLast seen
    prepare-releaseskillBump the version, date the Unreleased changelog section, commit as chore(release), tag, do not push32026-10-13
    conventional-commitsclaude-mdCommit messages use type(scope): subject; corrected in 3 projects32026-10-13
    ledgerlite-time-providermemoryledgerlite-api reads the clock through TimeProvider32026-10-09
    pebble-web-conventionsmemorypebble-web: date-fns not moment, no console.log, lint before commit32026-10-12

    Dropped (3): the docker compose → dotnet ef → dotnet test loop and the five shell recipes (how the assistant checks its work, not a choice you made), and "use the structured logger, not print" (seen in one session only).

    Draft: prepare-release/SKILL.md

    ---
name: prepare-release
description: Prepares a release in a repository by bumping the version in the project file, moving the Unreleased section of CHANGELOG.md under a new dated heading, committing as chore(release) and tagging, without pushing. Use when the user asks to prepare, cut or tag the next release.
---

# Prepare a release

1. Take the version from the request. If it is missing, ask for it.
2. Bump the version in the project file that carries it (`package.json`, `pyproject.toml`, `Directory.Build.props`, or the repository's equivalent).
3. In `CHANGELOG.md`, move the Unreleased section under a new heading with the version and today's date.
4. Commit the two files as `chore(release): <version>`.
5. Tag the commit with the version.
6. Do not push. Say that the commit and the tag are local.

    Next step

    Which of the four do you want to keep? Nothing is installed until you pick. Kept proposals are installed with /skill-miner:accept.

    in the project file, move the Unreleased section of CHANGELOG.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Repetitive manual corrections and recurring workflow steps drain developer productivity. When you ask it to, this skill scans your local Claude Code, Cursor and Codex transcripts to identify patterns, then proposes skills and project rules based on your actual behavior. Nothing is installed or removed without your pick.

    Use when you want to mine your own transcripts, review proposals, install the ones you approve, or prune generated skills that never trigger. Do not use it to read anyone else's transcripts or to summarize a single session.

    What it does

    • Pattern mining scans transcript history to find repeated multi-step procedures and corrections.
    • Skill generation drafts SKILL.md files for multi-step tasks you frequently perform manually.
    • Preference extraction identifies project-specific preferences and proposes text for your rules or memory files, which you approve before anything is written.
    • Conflict detection spots contradictions between your current habits and existing automated rules.
    • Context pruning identifies generated skills that are never triggered, preventing context bloat.

    How it works

    1. Scan runs the mining script to analyze transcripts from the last 30 days across supported IDEs and CLI tools.
    2. Review generates a proposal index where you can inspect drafted skills, rules, and their provenance.
    3. Install deploys approved proposals to the appropriate global or project-specific skill directories.
    4. Cleanup lists generated skills that never triggered and moves only the ones you pick to a recoverable folder.

    Frameworks & tools

    Built with Node.js (version 18 or higher). Interfaces with Claude Code, Cursor, and Codex transcript formats and skill directories.

    Why this beats prompting it yourself

    Manual prompt engineering for workflows requires remembering every edge case you hit last week. This skill extracts those specific constraints directly from your real-world usage data, ensuring automated rules reflect your actual requirements rather than idealized guesses.

    Use cases

    • Automate boilerplate sequences you keep typing in different projects.
    • Turn a preference you keep restating across projects into one rule you approve.
    • Remove obsolete automation that no longer aligns with your current workflow.
    • Spot where a rule you typed contradicts an existing skill or rule, with a proposed line saying which wins.

    Known limitations

    Transcripts skip tool outputs and injected context, focusing only on typed prompts and tool-call shapes. Compressed Codex rollouts require Node 22.15 or later for decompression. Cursor sessions carry no timestamps, so they are dated by file modification time. Installing and running the skill inside Codex and Cursor, and the optional live hooks, have not been tested in those tools.

    How to use

    Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill passes our trust checks below.

    Security scanned

    Passed our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scanned

    Verified clean 2 days ago

    • Free to download with an account

    Requires

    Node.js

    Listed2 days ago

    What's inside

    Frequently Asked Questions