agentsclimarketplace

Changelog

Skill couimet/my-claude-skills/skills/changelog

Create or update CHANGELOG entries with tone guardrails, thematic grouping, and implementation-detail leak detection.From its SKILL.md

Install
npx -y skills add couimet/my-claude-skills --skill changelog

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

10.4 KB, as published. Nobody here has run it

Changelog

Create a CHANGELOG entry with built-in guardrails for tone, placement, and detail-leak detection. Writes directly to CHANGELOG.md. The user reviews via git diff before committing.

Input: $ARGUMENTS (a short description of the change, e.g. "binding survives language-mode changes")

Step 1: Locate CHANGELOG.md

Find the target CHANGELOG.md using git diff to determine which part of the project has changes:

  1. Run git diff --name-only to see changed files
  2. If all changes are within a single package directory (e.g., packages/foo/), look for a CHANGELOG.md in that directory first
  3. If changes span multiple packages, or no package-level CHANGELOG.md exists, fall back to the root CHANGELOG.md
  4. If multiple package-level CHANGELOGs could apply (changes span packages that each have their own CHANGELOG), ask via AskUserQuestion which one to target

If no CHANGELOG.md exists anywhere, STOP: "No CHANGELOG.md found. Create one first."

Step 2: Read CHANGELOG Structure

Read the target CHANGELOG.md and the project's CLAUDE.md (if it exists) to understand:

  • Versioning scheme: detect from the file's own header or preamble. Common patterns:
    • SemVer: headings like ## [1.2.3] or ## [Unreleased], often with a preamble referencing semver.org or Keep a Changelog
    • CalVer: headings like ## 2026.03.19, often with a preamble referencing calver.org
    • Other: match whatever heading pattern the file uses
  • Category subsections: typically ### Added, ### Changed, ### Fixed, ### Removed (Keep a Changelog style). Some projects use different categories. Match what exists.
  • Entry format: how entries are written, how issue links are formatted, what style the project uses.
  • Existing entries: scan the most recent section(s) to understand the project's voice and level of detail.

Also check the project's CLAUDE.md for any explicit CHANGELOG conventions (heading format, link style, category rules). CLAUDE.md overrides inferred conventions.

Step 3: Detect Subsection

Auto-detect the most likely category from context:

SignalCategory
New files added (git diff --name-status shows A)Added
Bug fix (issue labels, commit messages mentioning "fix")Fixed
Deleted files or removed featuresRemoved
Everything elseChanged

If the detection is confident (clear signal from the table above), proceed without prompting.

If ambiguous (e.g., a file was both added and modified, or the change could be "Added" or "Changed"), confirm with AskUserQuestion:

AskUserQuestion(
  question: "Which CHANGELOG category for this entry?",
  header: "Category",
  options: [
    { label: "Changed (Recommended)", description: "Modifications to existing behavior" },
    { label: "Added", description: "New features or files" },
    { label: "Fixed", description: "Bug fixes" },
    { label: "Removed", description: "Deleted features" }
  ]
)

Reorder options so the auto-detected category appears first with "(Recommended)".

Step 4: Determine Version Heading and Position

Determine where to place the entry based on the versioning scheme detected in Step 2:

SemVer projects (with [Unreleased] section):

  • If an ## [Unreleased] section exists, place the entry there. This is the standard SemVer workflow for in-progress changes.
  • If no [Unreleased] section exists, create one above the most recent versioned heading.

CalVer projects (date-based headings):

  • If today's date heading already exists, check for micro suffixes. Create a new heading with the next micro suffix (e.g., 2026.03.19 exists → 2026.03.19.1; 2026.03.19.1 exists → 2026.03.19.2).
  • If today's date heading doesn't exist, create a new one. Use the date format matching existing headings (read from CLAUDE.md or infer from the file).

Other schemes:

  • Match the existing heading pattern. If uncertain, ask the user via AskUserQuestion.

Positioning Rule (All Schemes)

Newest entries go at the top. Version headings must appear in strict descending order — the most recent release at the top of the file, the oldest at the bottom. For CalVer: within the same calendar date, higher micro suffixes are newer and come first (e.g., 2026.07.14.2 above 2026.07.14.1 above 2026.07.14). For SemVer: within the same minor version, higher patch versions are newer and come first (e.g., [1.2.3] above [1.2.2] above [1.2.1]). When adding a new heading, insert it above all older headings — never append to the bottom. After writing, verify with grep '^## ' CHANGELOG.md | head -20. The output must print headings in descending order. If they are out of order, re-sort before presenting to the user.

Step 5: Draft the Entry

Write the entry following the tone rules (see Tone Rules section below). The entry should:

  1. Lead with what the user sees or experiences
  2. Be one sentence for the change, optionally one for previous behavior
  3. End with an issue reference following the project's link format (read from existing entries to match the convention)

Each bullet covers a single distinct change. When a PR or issue includes multiple unrelated changes, write separate bullets. Do not chain them together with periods or semicolons into a single entry. The "one sentence" rule is per change, not per category subsection.

Before drafting, read 3-5 recent entries in the CHANGELOG to match the project's voice and formatting conventions.

Step 6: Implementation-Detail Check

Scan the draft entry against the blocklist patterns (see Implementation-Detail Blocklist section). For each match:

  • Print a warning: Possible implementation detail: "<matched text>" Consider rephrasing in user-facing terms
  • Do NOT block. The user may have a legitimate reason to include the detail

If no matches, proceed silently.

Step 7: Thematic Placement

Scan the entries in the target section's category subsection for thematic proximity:

  1. Look for entries about the same feature, component, module, or user workflow
  2. If a match is found, place the new entry adjacent to the closest match (after it)
  3. If no match is found, append at the end of the subsection

When creating a new version heading, placement is straightforward. The entry is the first in its subsection.

Step 8: Present and Write

Show the user what will be written and where:

File: packages/foo/CHANGELOG.md
Category: ### Changed
Heading: ## [Unreleased]
Placement: after "existing related entry text..." (thematic match)
Entry:
- Search results now highlight matched terms ([#42](https://github.com/org/repo/issues/42))

Then write the entry to CHANGELOG.md using the Edit tool.

Tone Rules

These rules ensure entries describe what users experience, not how the code works internally.

Do

  • Lead with what the user sees: "Binding survives language-mode changes" not "Fixed onDidCloseTextDocument firing during language-mode switch"
  • Describe the symptom, not the mechanism: "no longer silently breaks" not "now checks isClosed flag before re-registering"
  • One sentence for the change, optionally one for previous behavior. No more
  • Use the feature name as the subject when the change is about a specific feature

Don't

  • Name internal APIs, event handlers, class names, or flags in the entry text
  • Explain the architecture or implementation approach
  • Include file paths or function names (the git diff shows those)
  • Write more than two sentences

Examples

Good (SemVer project, a task management app):

- Search results now highlight matched terms in the preview pane ([#42](https://github.com/acme/taskflow/issues/42))

Bad (same change, leaking implementation):

- Added ElasticSearch highlight fragments to SearchResultDTO and mapped them through the PreviewRenderer pipeline with dangerouslySetInnerHTML for term emphasis ([#42](https://github.com/acme/taskflow/issues/42))

Good (CalVer project, a CLI tool):

- `deploy` command now retries on transient network errors instead of failing immediately ([issues/87](https://github.com/acme/shipit/issues/87))

Bad (same change, leaking implementation):

- Wrapped the HTTP client's POST call in a retry loop with exponential backoff using the got library's retry option with a maxRetryAfter of 30000ms and calculateDelay hook ([issues/87](https://github.com/acme/shipit/issues/87))

Implementation-Detail Blocklist

Flag entries containing these patterns. Emit a warning and allow the entry. Rare edge cases may legitimately include a technical name.

PatternWhy it's flagged
onDid, onWillEvent handler names (VS Code, etc.)
setTimeout, setInterval, requestAnimationFrameTimer/scheduling implementation details
mock, stub, spyTest infrastructure, not user-facing
listener, handler, callback, middlewareInternal architecture
PascalCase words that aren't feature namesLikely class/type names (e.g., SearchResultDTO, PreviewRenderer)
flag, isFoo, hasFooInternal boolean state
Event name patterns (Event, Emitter, dispatch)Internal pub/sub details
Library-internal APIs (dangerouslySetInnerHTML, useEffect, __dirname)Framework internals

Thematic Grouping Rules

Before placing an entry, scan the target subsection for thematic proximity:

  1. Same feature or command name: entries mentioning the same feature, command, or module name are the strongest match
  2. Same user workflow: entries about the same user-facing workflow (e.g., "search", "deploy", "authentication")
  3. Same component area: entries about the same part of the system (e.g., "CI", "build", "tests", "config")

Place the new entry after the last thematic match in the subsection. If no match exists, append at the end.

Formatting

See /prose-style for hard-wrap and GitHub-reference rules.

Before writing the entry, skim the text for AI-writing tells: em dashes, filler phrases (in order to, due to the fact that), vague attributions, generic positive conclusions. Rewrite any you find.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.