fusion-package-scribe
Skill2 starsChanged 2 months agoMITView on GitHub
Systematically improves TSDoc and README documentation across packages in a TypeScript monorepo. Discovers public API surfaces, generates or improves TSDoc on all exports, rewrites READMEs to a consistent retrieval-friendly structure, and commits each package independently. USE FOR: monorepo-wide documentation passes, single-package doc improvements, TSDoc generation on public exports, README standardization, review council quality checks. DO NOT USE FOR: runtime code changes, security scanning, API reference site generation, non-TypeScript languages.
In this repository31 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .github/copilot-instructions.mdCopilot instructions
- caveman-compressSkill
- fusion-core-servicesSkill
- fusion-dependency-reviewSkill
- fusion-designSkill
- fusion-framework-mockingSkill
- fusion-framework-testingSkill
- fusion-github-review-resolutionSkill
- fusion-issue-solvingSkill
- fusion-issue-task-planningSkill
- fusion-mcpSkill
- fusion-pm-mcpSkill
- fusion-backend-devSkill
- fusion-developer-appSkill
- fusion-developer-portalSkill
- fusion-devtoolsSkill
- fusion-help-apiSkill
- fusion-help-docsSkill
- fusion-help-integrationSkill
- fusion-infra-cliSkill
- fusion-issue-authoringSkill
- fusion-roles-cliSkill
- fusion-rulesSkill
- fusion-skill-authoringSkill
- fusionSkill
- fusion-skillsSkill
- equinor-design-systemSkill
- fusion-code-conventionsSkill
- fusion-researchSkill
- fusion-rule-authorSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
30 rules. About 0 are common, 22 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈22
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 34
Use when you need to systematically add or improve documentation across TypeScript packages — whether a full monorepo sweep or a single …
- Distinctive: line 60
If required inputs are missing or ambiguous, ask before proceeding.
- Distinctive: line 106
Before generating any documentation:
- Distinctive: line 108
Search for repo-level documentation instructions:
- Very specific: line 111
Read tsconfig.json to understand module structure, path aliases, and strict mode settings
- Very specific: line 112
Read biome.json or equivalent linter config for style expectations
- Very specific: line 113
Check if fusion-code-conventions is available — if so, defer to its TSDoc rules from references/typescript.conventions.md
- Very specific: line 114
If no repo-level standards exist, use references/tsdoc-checklist.md as the quality baseline
- Very specific: line 120
Read the root package.json or workspace config (pnpm-workspace.yaml, lerna.json, turbo.json) to find all packages
- Very specific: line 123
Write the discovery summary to .tmp/scribe-discovery-<context>.md
- Distinctive: line 132
If a single package exceeds 60% of the estimated context window, flag it for special handling (barrel exports first, defer internal modules)
- Very specific: line 133
Write the execution plan to .tmp/scribe-plan-<context>.md
- Distinctive: line 150
Read the existing implementation to understand intent, parameters, return values, error paths, and side effects
- Distinctive: line 151
Check for existing TSDoc — improve rather than replace when present
- Distinctive: line 161
Do not modify runtime code — only doc comments
- Distinctive: line 165
Read the existing README (if any) to preserve valuable content
- Distinctive: line 173
Ensure the README is useful to a developer discovering the package for the first time
- Distinctive: line 189
After the review council passes for a package:
- Distinctive: line 192
Commit using the repo's conventional commit format, defaulting to:
- Distinctive: line 196
Do not push — leave that to the user or a PR workflow
- Distinctive: line 200
After all packages are processed:
- Very specific: line 202
Write a summary to .tmp/scribe-report-<context>.md:
- Distinctive: line 208
If a tracking issue was provided, update it with the summary
- Distinctive: line 222
Only modify doc comments and README files — never touch runtime code
- Distinctive: line 223
Do not push commits; leave push decisions to the user
- Distinctive: line 224
Do not auto-merge PRs or bypass review workflows
- Distinctive: line 225
Do not invent API behavior — document only what the code actually does
- Distinctive: line 226
Do not delete or replace existing valuable documentation without preserving its content
- Distinctive: line 227
When a tracking issue is referenced, update it only after explicit confirmation
- Distinctive: line 229
If the review council fails a package twice, flag it for manual review instead of looping indefinitely
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Package Scribe
Experimental caveat
This skill is experimental and not yet stable. Behavior, structure, and outputs may change between versions.
When to use
Use when you need to systematically add or improve documentation across TypeScript packages — whether a full monorepo sweep or a single package.
Typical triggers:
- "document all packages"
- "improve docs for
packages/utils/observable" - "add TSDoc to all public exports in this package"
- "rewrite the README for this package"
- "run a documentation pass on the monorepo"
- "review the TSDoc quality in this package"
Implicit triggers:
- A package has public exports with missing or stub TSDoc
- A package README is outdated, inconsistent, or missing standard sections
- A documentation sweep is planned across multiple packages
When not to use
- Modifying runtime code (only doc comments and README files)
- Generating API reference sites or TypeDoc output
- Non-TypeScript languages (initial scope is TS/TSX only)
- Auto-merging PRs or bypassing review workflows
- Security vulnerability scanning or performance profiling
- Replacing existing CI-based doc generation pipelines
Required inputs
If required inputs are missing or ambiguous, ask before proceeding.
- Scope: monorepo root path or specific package path(s)
- Mode:
single(one package) orsweep(multiple packages)
Conditional inputs
- Package filter: glob or list when processing a subset of packages
- README template path: repo-specific template if not using the built-in default (for example under
.github/instructions/) - Tracking issue: issue number for lifecycle updates (assign, status, close)
- Commit style: repo-specific conventional commit format (defaults to
docs(<package>):) - Batch size: number of packages per batch for sweep mode (defaults to 5)
Defaults
- Mode:
single(if only one package path provided) - Commit prefix:
docs(<package-name>): - Batch size: 5 packages per sweep batch
- README structure: built-in template from
references/readme-template.md - Review council: enabled (runs after every package)
Precedence and standards discovery
This skill discovers and follows the target repository's own standards:
- Repository instructions — applicable files under
.github/instructions/,CONTRIBUTING.md,contribute/ - Tooling configuration —
tsconfig.json,biome.json,.editorconfig - Companion skill —
fusion-code-conventionsfor TSDoc rules, naming conventions, and intent quality (when installed) - Built-in defaults —
references/tsdoc-checklist.mdandreferences/readme-template.md
Repository-level standards always win. When no repo standards exist, the built-in defaults apply.
Agent modes
| Agent | Role | Activated for |
agents/orchestrator.agent.md | Batch coordinator | Sweep mode — plans batches, manages token budgets, tracks progress |
agents/documenter.agent.md | Per-package writer | Every package — scans API surface, generates TSDoc, rewrites README |
agents/reviewer.agent.md | Review council | After each package — verifies intent, comprehension, retrieval fitness |
In single-agent runtimes, all three roles run inline sequentially.
Instructions
Step 1 — Discover repository standards
Before generating any documentation:
- Search for repo-level documentation instructions:
- applicable files under
.github/instructions/ CONTRIBUTING.md,contribute/
- applicable files under
- Read
tsconfig.jsonto understand module structure, path aliases, and strict mode settings - Read
biome.jsonor equivalent linter config for style expectations - Check if
fusion-code-conventionsis available — if so, defer to its TSDoc rules fromreferences/typescript.conventions.md - If no repo-level standards exist, use
references/tsdoc-checklist.mdas the quality baseline
Step 2 — Discover packages
- For single mode: validate the provided package path exists and has TypeScript source files
- For sweep mode:
- Read the root
package.jsonor workspace config (pnpm-workspace.yaml,lerna.json,turbo.json) to find all packages - Filter by any provided glob/list
- Sort packages by estimated size (file count) for batch planning
- Write the discovery summary to
.tmp/scribe-discovery-<context>.md
- Read the root
Step 3 — Plan execution (sweep mode only)
Activate agents/orchestrator.agent.md (or run inline):
- Group packages into batches of the configured batch size
- Order batches: smaller packages (fewer source files) first to maximize early throughput
- Estimate token budget per package: ~100 tokens per source file for reading, ~200 tokens per export for TSDoc generation
- If a single package exceeds 60% of the estimated context window, flag it for special handling (barrel exports first, defer internal modules)
- Write the execution plan to
.tmp/scribe-plan-<context>.md
Step 4 — Process each package
Activate agents/documenter.agent.md (or run inline) for each package:
4a — Scan public API surface
- Find the barrel export file (
index.ts,index.tsx, ormainfield inpackage.json) - Trace all re-exports to identify the full public API surface
- Categorize exports: functions, classes, types/interfaces, constants, hooks, enums
- Prioritize: barrel exports first, then direct public exports, then internally-consumed-but-exported items
4b — Generate or improve TSDoc
For each public export:
- Read the existing implementation to understand intent, parameters, return values, error paths, and side effects
- Check for existing TSDoc — improve rather than replace when present
- Apply the TSDoc checklist (repo standards or
references/tsdoc-checklist.md):- Summary line: explain why and what problem it solves, not just what it does
@paramfor every parameter with meaningful descriptions@returnsfor every non-void function@templatefor every generic type parameter@throwsfor meaningful error paths@examplefor user-facing and non-trivial public APIs@deprecatedwith replacement guidance when applicable
- Flag and rewrite "name-echo" patterns (for example
/** Gets the value. */ getValue()) - Do not modify runtime code — only doc comments
4c — Rewrite or improve README
- Read the existing README (if any) to preserve valuable content
- Apply the README structure from repo instructions or
references/readme-template.md:- Package name and description
- Features / key exports
- Installation
- Usage with code examples
- API reference (summary of key exports with links or inline docs)
- Configuration (if applicable)
- Ensure the README is useful to a developer discovering the package for the first time
- Optimize for retrieval: use clear headings, keyword-rich descriptions, and concrete examples
Step 5 — Review council
Activate agents/reviewer.agent.md (or run inline) after each package:
- Intent extraction — Does the TSDoc accurately describe what the code does and why? Flag any comment that merely restates the function/type name.
- Code comprehension — Are complex algorithms, state machines, or side effects explained? Would a new developer understand the code from the docs alone?
- User-facing quality — Is the README useful to someone discovering the package for the first time? Does it have working examples?
- Retrieval fitness — Will the documentation produce good hits in RAG / semantic search? Are key terms present in headings and summaries?
The reviewer produces a pass/fail per criterion. Failures loop back to Step 4 for the specific package.
Step 6 — Commit
After the review council passes for a package:
- Stage only documentation files (
.ts/.tsxfiles for TSDoc changes,README.md) - Commit using the repo's conventional commit format, defaulting to:
docs(<package-name>): improve TSDoc and README documentation - Do not push — leave that to the user or a PR workflow
Step 7 — Report
After all packages are processed:
- Write a summary to
.tmp/scribe-report-<context>.md:- Packages processed and status (pass/fail/skipped)
- Total exports documented
- Review council pass rates per criterion
- Any packages flagged for manual review
- Commits created
- If a tracking issue was provided, update it with the summary
Expected output
- TSDoc comments on all public exports in processed packages
- Consistent README for each processed package
- One commit per package with documentation-only changes
- Summary report in
.tmp/ - Tracking issue update (when issue number provided)
Safety & constraints
This skill is mutation-capable. Repository-local workflow instructions take precedence over inline guidance when they conflict.
- Only modify doc comments and README files — never touch runtime code
- Do not push commits; leave push decisions to the user
- Do not auto-merge PRs or bypass review workflows
- Do not invent API behavior — document only what the code actually does
- Do not delete or replace existing valuable documentation without preserving its content
- When a tracking issue is referenced, update it only after explicit confirmation
- Respect
.gitignoreand do not commit temporary files from.tmp/ - If the review council fails a package twice, flag it for manual review instead of looping indefinitely
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
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.

