More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Release Notes from Commits: Draft and flag bad messages

    1

    Turn a commit log into a changelog draft and see which messages broke the format first. Groups conventional commits, calls out breaking changes and lists non-conventional ones for you to fix.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Release Notes from Commits: Draft and flag bad messages

    Example session with this skill installed

    Review references/bad.json after v1.4.2, dated 2026-11-01. Use strict mode and show what prevents accepting the proposed version.

    • Read your context and instructions
    • Compiled the release notes from

    Actual stdout from `python3 -I -B scripts/release_notes.py --input references/bad.json --current-version v1.4.2 --date 2026-11-01 --strict` (exit 1): # Release review - **Version:** v1.4.2 -> **1.4.3** (bump: patch) - **Commits read:** 2; changelog entries: 1; breaking: 0; non-conforming: 1; dropped by revert pairing: 0 - **Bump driven by:** aaaaaa1 (fix) ## Not Conventional Commits (not in the notes; triage these) - `bbbbbb2` ship the new format -- has a BREAKING CHANGE footer, check it ## Changelog section ## [1.4.3] - 2026-11-01 ### Fixed - preserve empty columns (aaaaaa1) What the agent tells you: blocked until you look at `bbbbbb2`. The one valid fix points to 1.4.3, but `bbbbbb2` claims a breaking change and its header didn't parse, so it never counted. "breaking: 0" only covers the commits that parsed. Check what `bbbbbb2` changed and whether users need to migrate, fix or confirm the message, and rerun. The fix entry stays as a provisional draft. No files were written.

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

    About this skill

    **The problem** It's release day and you have a few hundred commits since the last tag. Most are chores and docs, a few matter to users, and one or two don't follow your commit convention at all. Changelog generators quietly skip those odd ones. That's a problem when one of them says "BREAKING CHANGE" in its body and your tool still suggests a patch release. **See it before you buy: real input, real output** Command (from the package folder, exit code 1, 0.024 s): ``` python3 -I -B scripts/release_notes.py --input references/bad.json --current-version v1.4.2 --date 2026-11-01 --strict ``` *Before: Commit export (references/bad.json)* ``` [ { "hash": "bbbbbb2", "message": "ship the new format\n\nBREAKING CHANGE: legacy requests require migration." }, { "hash": "aaaaaa1", "message": "fix: preserve empty columns" } ] ``` *After: lines copied from the script's output (excerpt):* ``` - **Version:** v1.4.2 -> **1.4.3** (bump: patch) - **Commits read:** 2; changelog entries: 1; breaking: 0; non-conforming: 1; dropped by revert pairing: 0 ## Not Conventional Commits (not in the notes; triage these) - `bbbbbb2` ship the new format -- has a BREAKING CHANGE footer, check it ## [1.4.3] - 2026-11-01 - preserve empty columns (aaaaaa1) ``` **What the numbers show** - Read 803 commits from 5 public projects at pinned commits, producing 195 changelog entries; 40 sampled headers and all 21 breaking notes were checked by hand against the original messages. - With the date fixed, all 25 runs gave byte-identical output, and all 5 --strict runs exited 1 because unparsed messages were present. - Honest limit: three of the five exports are single-commit shallow clones, so they show triage works, not that the notes read well. **What you get** - One Python script (standard library only) plus instructions the agent follows to review the draft with you. - It reads full commit messages, either as a JSON array or as a `git log` export with separators, so breaking-change footers in the body aren't lost. - Three output styles: a review report, a plain Markdown changelog section, or JSON. - The usual Keep a Changelog headings (Added, Changed, Deprecated, Removed, Fixed, Security). Chores, docs, CI and similar commits are hidden unless you ask for them. - A proposed next version with the commits that drove it. You pick how 0.x versions bump (three choices). - A revert and the commit it undoes cancel out when both are in the range. If it can't tell which commit a revert targets, it stops instead of guessing. - A list of messages it couldn't parse. `--strict` turns that list into exit code 1 for CI, and the draft still prints. The agent asks for your release range and current version, runs the script, then walks you through breaking changes, hidden commits and unparsed messages. You get a provisional version, a draft section, and a short list of questions only you can answer. **Measured on real data (full detail)** I ran the final script on commit exports from 5 public projects, pinned to fixed commits: 803 commits in total, which produced 195 changelog entries. Two of the exports have 400 commits each. The other three are shallow clones with a single non-conventional commit each, so they only show that triage works, not that release notes come out right. | Repository | Pinned commit prefix | Commits read | Accepted conventional headers | Breaking notes | Median ms | | --- | --- | ---: | ---: | ---: | ---: | | conventional-changelog | 280319e792ce | 400 | 397 | 21 | 34.618 | | commitizen | 33c248e4d21f | 400 | 398 | 0 | 29.731 | | flask | 49943b5a9441 | 1 | 0 | 0 | 28.352 | | click | 2247b35ea1c4 | 1 | 0 | 0 | 27.131 | | requests | 3c7baaa05ab4 | 1 | 0 | 0 | 33.563 | "Accepted conventional headers" means the subject line parsed as `type(scope): description`. Times are the median of 5 runs and include Python startup (Python 3.13.5, Linux). With the date fixed, each of the 25 runs gave byte-identical output for its input, and all 5 `--strict` runs exited 1, as they should with unparsed messages present. I also checked 40 sampled headers and all 21 breaking notes by hand against the original messages. Every one matched the documented rules. That shows the script reads messages correctly, not that it catches API breaks nobody declared. Ten of the 21 came from a `!` in the header with no recognised footer, so the note is just the commit subject. Eight of those ten had their explanation under a plural `BREAKING CHANGES:` heading, which the script doesn't read, so you'll want to copy that text in by hand. Three malformed headers had breaking footers, and those stay in the triage list rather than the draft. Pins, commands and raw timings are in `references/benchmarks.md`. **Known limitations** - Messages that don't parse are left out of the draft and the version. The plain Markdown output doesn't show them, so read the report or JSON first. - It only knows what commit messages say. It can't spot an API break nobody declared, and it won't make up migration steps. - It follows the spec's singular `BREAKING CHANGE:` footer. A body that says `BREAKING CHANGES:` (plural) without a `!` in the header isn't treated as breaking. - A revert is assumed to undo the whole commit. There's no per-package versioning for monorepos, and it doesn't bump prerelease numbers like `-rc.1` to `-rc.2`. - It loads the whole export into memory, and revert matching gets slower as history grows (it's quadratic). The timings above say nothing about very large histories. - Commit text, links and HTML pass through as-is, so check them before you publish. It never tags, publishes, or edits your CHANGELOG file. **Who it's for** Maintainers and release engineers whose team mostly writes conventional commit messages and who want a repeatable draft that shows what it left out. **Requires** Python 3.8+ and a UTF-8 export of full commit messages, newest first. You only need Git to make the export. The script is a single file and runs fine copied to a temp folder. Privacy: runs locally with no network, no telemetry, and no file writes unless you redirect output. Nothing is sent anywhere. **Check it yourself** 33 test cases pass, and 6 saved sample outputs match the script byte for byte. The script also ran on 803 real commits from 5 public projects (25 timed runs plus 5 strict runs). This version was trimmed to fit under 20 KB, and 49 test inputs gave exactly the same output and exit codes before and after. To check it yourself from the package folder, run `python3 -I -B scripts/release_notes.py --input references/bad.json --current-version v1.4.2 --date 2026-11-01 --strict`, expect exit 1, and compare the output with `references/sample-output/bad-report.md`. The full check command is in `references/specifications.md`, and the real-history commands are in `references/benchmarks.md`. **Price and guarantee** $9 one time, yours to use with any agent that can run Python. Covered by Agensi's money-back guarantee if it doesn't do what this page promises.

    Requires full commit bodies, not just subjects. Does not analyze source code for hidden API breaks. Limited to Python 3 environments. No support for multi-package monorepo versioning.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 15 days ago

    Requires

    Git

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed15 days ago
    Updated5 days ago

    What's inside

    Frequently Asked Questions

    If they already work for you, keep using them. They're good at writing the changelog. This is for the review step before you publish: it lists what it skipped, flags breaking changes in messages that didn't parse, and has the agent ask you about them instead of quietly bumping a patch.

    They're listed for triage in the report and JSON, and they don't affect the draft or the version. With `--strict` the script exits 1 but still prints everything. Git's own `Revert "..."` and `Reapply "..."` subjects are understood.

    No. You give it an export of the commits you want covered. Exporting, writing to your CHANGELOG, tagging and publishing are all up to you.

    Agensi's money-back guarantee covers every paid skill: if it doesn't do what this page promises, you get your money back. Everything above is output from the script in the package, so you can see what it does before you buy.

    New in Technical Documentation