@riekelt/writing-changelogs

Use when writing a changelog entry, release notes, or a "what shipped" summary after completing work. Encodes the entry shape, the handover template, and the honesty conventions. Use after shipping meaningful work, even if the user just says "summarize what we did".

View in AI SkillSafe app
0 downloads
0 stars
0 demos
SKILL.md

Technical writer

skills.sh

Skills for writing technical documents in a strict house style: conclusion first, every claim traceable to a source, one fact in one home, and a banned-constructions list that catches machine text. Distilled from writing conventions used across my own repositories.

One core skill holds the shared rules; ten specialized skills build on it.

Skill Use when
technical-writing Any technical document. The foundation: read-first, doc classification, hard rules, audience, the pre-draft checkpoint. Carries style.md (sentences, words, headings, banned constructions) and truth.md (claims, sourcing, confidence, staleness).
writing-design-docs Proposals, RFCs, design docs, specs, migration plans. Skeleton, Why & What box, completeness checks.
recording-decisions ADRs and decision log entries.
writing-changelogs Changelog entries and release notes.
writing-runbooks Runbooks, setup guides, troubleshooting, procedures.
writing-issues Tracker items: epics, stories, tasks, bug reports, spikes, acceptance criteria.
writing-postmortems Postmortems, incident reports, root-cause analyses, near misses.
documenting-legacy-codebases Documenting inherited or under-documented systems: survey-first inventory, evidence hierarchy, dead-or-alive proofs, findings register, coverage ledger, and the parallel grounding campaign.
diagramming-processes Process, lifecycle, and interaction diagrams: diagrams as source, the ArchiMate-to-PlantUML notation ladder, behavior-level participants, the diagram index.
documenting-contracts API references: the four detail levels, the DTO-first payload catalog, wire types with source anchors, omitted-versus-null semantics.
reviewing-technical-prose Reviewing or rewriting someone else's text; severity mapping; the delivery checklist.

Influences and prior art

The rules are house-composed, but most stand on named public constructs. What each one contributed, and where this skill set deliberately diverges:

Construct What it contributed Where it landed
Nygard ADRs and MADR The decision record format: context, decision, consequences, alternatives; immutability once accepted recording-decisions. Divergence: the Negative consequences section is mandatory and may not be empty
Keep a Changelog Categories, newest-first ordering, one entry per change writing-changelogs. Divergence: entries also log known issues, deferred items, and deliberate omissions, which generated changelogs cannot know
ASD-STE100 simplified technical English One instruction per sentence, sentence-length ceilings, controlled-language procedure mechanics references/style.md sentence rules, writing-runbooks
Zinsser, On Writing Well Plain-prose discipline: active voice, cut clutter, one term per concept references/style.md
Google developer documentation style guide Heading types: noun phrases for concepts, task headings for procedures; sentence-case headings references/style.md headings
digital.gov plain language The case against question headings outside real FAQs references/style.md headings
Wikipedia: Signs of AI writing and humanizer-style catalogs The machine-tell inventory: negative parallelism, rule of three, importance announcements, cursed vocabulary references/style.md banned constructions. Divergence: scoped to technical documents, coupled to review severities, and paired with a what-not-to-flag list to protect human text
Diátaxis The insight that document types must not mix The classification table in technical-writing. Divergence: the kinds here are keyed to the edit rule (normative, descriptive, historical, runbook, reference), not to reader need
BLUF (bottom line up front) Conclusion first at every level technical-writing workflow, the design-doc summary
Conventional Commits and semantic-release Commit subjects as the changelog, automated versioning The release pipeline of this repo
multi-agent-review The plugin and marketplace repository structure, and the severity-tagged finding format with a named empty case The repo layout, reviewing-technical-prose

No prior art turned up for these parts when I searched public agent skills, style guides, and tooling in August 2026:

  • the per-kind edit semantics (never water a normative doc down to match violating code)
  • claim provenance with confidence tiers
  • the staleness rules
  • one fact one home, with "the source wins and the index is the bug"
  • the rewrite rule that an added fact counts as an error like a lost one
  • the remove-the-name test

Agents and commands

Beyond the skills, the plugin ships two subagent definitions and two commands (Claude Code; other platforms load what they support):

Component What it does
prose-reviewer agent The fresh-eyes review protocol as an agent type: digest-bound, cross-references followed, findings in the severity grammar, never edits.
doc-grounder agent Read-only grounding of one document against the code at HEAD, under the legacy-campaign doctrine.
/prose-review [file ...] Runs a digest-bound prose-reviewer per file and relays the verdicts.
/ground-docs [path] Launches the legacy documentation campaign: survey, plan for approval, fan-out, ledger.

Install

Claude Code:

/plugin marketplace add riekelt/technical-writer
/plugin install technical-writer@technical-writer

Other agents: point the platform's plugin loader at plugins/technical-writer/, or symlink the directories under plugins/technical-writer/skills/ into the agent's skills directory.

Repository layout

.claude-plugin/marketplace.json          # Claude Code marketplace manifest
.agents/plugins/marketplace.json         # generic agents marketplace manifest
.github/workflows/release.yml            # semantic-release on push to main
.releaserc.json                          # release config; stamps versions into package.json and the plugin manifests
plugins/technical-writer/
  .claude-plugin/plugin.json             # Claude Code plugin manifest
  .codex-plugin/plugin.json              # Codex plugin manifest
  .cursor-plugin/plugin.json             # Cursor plugin manifest
  evals/                                 # skill trigger and behavior evals
  agents/                                # prose-reviewer, doc-grounder subagent definitions
  commands/                              # /prose-review, /ground-docs entry points
  skills/
    technical-writing/                   # core: SKILL.md, references/style.md, references/truth.md
    writing-design-docs/
    recording-decisions/
    writing-changelogs/
    writing-runbooks/
    writing-issues/
    writing-postmortems/
    documenting-legacy-codebases/
    diagramming-processes/
    documenting-contracts/
    reviewing-technical-prose/

Releases

Conventional Commits on main drive semantic-release: commit subjects become the changelog, and the release stamps the version into package.json and all three plugin manifests. CHANGELOG.md is generated; do not hand-edit it.

License

MIT

Embed badges

Add these to your README to show the skill's verification status.

SkillSafe verified badge
Verified badge
[![SkillSafe verified badge](https://api.skillsafe.ai/v1/badge/@riekelt/writing-changelogs/verified)](https://skillsafe.ai/skill/@riekelt/writing-changelogs/)
Installs badge
Installs badge
[![Installs badge](https://api.skillsafe.ai/v1/badge/@riekelt/writing-changelogs/installs)](https://skillsafe.ai/skill/@riekelt/writing-changelogs/)
Scan badge
Scan badge
[![Scan badge](https://api.skillsafe.ai/v1/badge/@riekelt/writing-changelogs/scan)](https://skillsafe.ai/skill/@riekelt/writing-changelogs/)
Eval pass rate badge
Eval pass rate
[![Eval pass rate badge](https://api.skillsafe.ai/v1/badge/@riekelt/writing-changelogs/eval)](https://skillsafe.ai/skill/@riekelt/writing-changelogs/)