dev-docs
Skill0 starsChanged 2 months agoMITView on GitHub
Maintain product and technical documentation, requirements, lean work queues, decisions, and verification without duplicating contracts. Use for product grooming, references and guides, recovering requirements from code, bugs, research, change-impact and consistency checks. Reconcile selected existing sources into owning requirements and work records when their semantics do not yet fit this model. Preserve compatible documents and capture new input directly in its owners. Use migrate for an explicitly requested one-shot transition.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- Linchi1997yeh/craftsman-suitetheir own version, first seen Oct 5, 2026
- taiga-shiokawa/claude-skillstheir own version, first seen Sep 2, 2026
- amjadjibon/skillstheir own version, first seen Jul 12, 2026
- lilyjem/dev-docs-skilltheir own version, first seen Jan 22, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- amjadjibon/skills★ 2same name
- Linchi1997yeh/craftsman-suite★ 2same name
- lilyjem/dev-docs-skill★ 1same name
- taiga-shiokawa/claude-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
- cost-auditSkill
- dev-rulesSkill
- doc-researchSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 8 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Keep the existing queue/index as the shared selection route for old and new work.
- Very specific: line 36
Use requirement IDs and existing layer definitions: new layer-based requirements use R-<LAYER>-NNN, numbered separately from existing work.
- Distinctive: line 40
Read applicable project instructions and the selected canonical skill source.
- Distinctive: line 50
Read-only requests produce a proposal only.
- Distinctive: line 54
Do not migrate unrelated linked history merely because it was encountered.
- Distinctive: line 57
Do not invent accepted requirements, owners, constraints, implementation evidence, approval, or missing historical facts.
- Distinctive: line 99
Do not load the entire backlog, all references, or every linked document for one change.
- Distinctive: line 100
Use the affected area and its applicable constraints.
- Distinctive: line 117
Create only needed documents.
- Distinctive: line 140
Check affected links/anchors and, where those records exist, ID uniqueness, dependencies, canonical queue ownership, conversion/retirement …
≈ “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
dev-docs
Default model and first action
The current work model is requirements-driven: specifications own obligations, queues request work, and results record delivery/verification. Descriptions, API references, guides, and runbooks maintain product/system knowledge through their own useful reading routes. Determine each project's source formats and authoritative owners from actual material; no particular task or document scheme is assumed to be obsolete. Keep the existing queue/index as the shared selection route for old and new work; use existing queues during transition. Queues are linked ID/name/status rows; context and results stay at their linked owners. Tasks own execution; requirements own contracts. A task contains its ID/action, links to exact requirements/revisions or other grounds, execution status, actual task waits, and result links. Keep context in canonical requirements, decisions, defects, research, or results, not in the task. Queue rows repeat task status, never requirement acceptance or verification state. Read the task and its linked grounds before execution. Requirement revisions identify contract versions; work needed to deliver them is captured in separate tasks in the same change. Preserve completed tasks and their checked baselines.
Register useful outcomes under the queue rules, not the
supporting record maintenance performed during the current work.
Use requirement IDs and existing layer definitions:
new layer-based requirements use R-<LAYER>-NNN, numbered separately from existing work.
Preserve existing stable IDs unless normalization is requested; explicit project overrides apply.
- Read applicable project instructions and the selected canonical skill source. Respect explicit project/user constraints, existing paths, language, and accepted contracts.
- Identify the reader's question and material type using documentation purpose and ownership. Reuse its owning document; requirements and queue entries are created only when their meanings and registration rules apply.
- Identify the selected work and affected contracts. Before changing or executing selected existing material whose contract/work ownership does not fit this model, follow on-touch conversion. Do this within the authorized work; no init, full-backlog migration, or new permission ceremony is needed. Read-only requests produce a proposal only. An explicit prohibition on migration or a conflicting binding project instruction must be surfaced, not silently bypassed.
- Continue from the owning records for the selected scope. Reuse canonical documents and conversion mappings before creating anything. Do not migrate unrelated linked history merely because it was encountered. Resolve current contracts, proposals, and history before treating found text as binding; follow retirement/replacement notices.
- Do not invent accepted requirements, owners, constraints, implementation evidence, approval, or missing historical facts. Preserve unknowns as questions. Capture material new input under ongoing capture; agreed changes revise the current requirement. Recover undocumented behavior/constraints through code recovery, with explicit provenance. Verify coverage before claiming any selected existing range transferred without loss.
This entrypoint, forms, bootstrap, and migration references all use the new model. Source-skill changes do not automatically rewrite an installed pinned copy or every consuming project.
Read only what the work needs
- Documentation types, canonical ownership, decision approval, work-context placement, proportionality, and granularity: documentation governance.
- Requirements, specification files, stable IDs, revisions, rationale, applicability, implementation prerequisites, proposal/retirement lifecycle, conflicts, and impact analysis: requirements.
- Work selection, readiness, cross-queue waits, research and non-code results, completion/history, and delivery targets: work queues.
- Existing material that needs semantic reconciliation: source assessment and on-touch conversion.
- New requirements, clarifications, and material questions: ongoing capture.
- Finding and recording undocumented requirements from implementation: code recovery and provenance.
- Grooming and incremental product development: product discovery.
- Bugs, findings, ideas, severity, and releases: maintenance.
- Source-heavy factual questions: bounded research handoff.
- Explicit one-shot migration: migration.
- Adopting/updating the skill as project governance: adoption and bootstrap.
- Optional examples: forms.
- An existing parser that actually constrains the work: validator compatibility.
Do not load the entire backlog, all references, or every linked document for one change. Use the affected area and its applicable constraints; follow dependencies/consumers only where they affect the result. Compact reading must still preserve cross-cutting contracts.
Ordinary work and actions
Ordinary maintenance updates the material that answers the selected question. Capture applicable obligations or convert selected existing scope when needed; use the queue's registration threshold for independently tracked work. A local documentation edit within existing work does not need another R/Q record. Preserve compatible conventions without requiring adoption or a whole-repository reorganization.
- discover — refine an idea/current product into draft, proposed, or accepted requirements, open questions, and the nearest useful deliverable. Groom specifications and prepare work separately; unknown implementation can become research without changing an accepted product requirement to draft. Respect discussion/read-only mode.
- init — when explicitly requested, adopt the needed documentation entrypoints through a complete project-local skill copy or a pinned public source. Create only needed documents; reuse existing layer definitions instead of creating an empty registry/template tree or parallel tracker.
- audit — read-only consistency and impact review of the selected scope. Classify descriptions, guides/procedures, specifications, decisions, queues, results, and source records by meaning, regardless of directory. Show conflicting formulations, applicability, consequences, and proposed resolutions. Do not certify factual claims or product behavior without the necessary source-grounded research/verification.
- migrate — explicitly convert the entire agreed scope in one controlled operation using the migration workflow. Preserve semantics, acceptance sources, history, and recoverability; reconcile all in-scope links and work. Destructive cleanup requires explicit authorization.
- update — compare the selected skill revision, update affected governance consumers and forms, and retire conflicting old routing under adoption. Installing/updating rules does not imply a full backlog conversion: remaining source records follow source assessment and the on-touch path when needed.
Completion and boundaries
Use closure reconciliation for the selected result and material type. Work readiness, execution briefs, completion, and history follow work queues; requirement revisions and applicability follow requirements.
Check affected links/anchors and, where those records exist, ID uniqueness, dependencies, canonical queue ownership, conversion/retirement redirects, proposal separation, and current versus historical applicability. Successful lint does not prove semantic compatibility. State unconverted scope, unresolved conflicts, missing sources, and verification limits honestly.
Git, worktrees, agents, CI, PRs, release, and production delivery follow the active project-work policy. This skill defines what must be captured and verified; it does not mandate PR-based delivery, install a runner, or grant merge/deploy authority. Code, documents, datasets, research, configurations, and external actions can all be outputs; publication and completion depend on the agreed target and actual confirmation.
The package provides Markdown guidance and examples. It does not claim that indexes, backlinks, conflict discovery, or synchronization are automated unless a project actually implements them. Do not create a new tracker/runtime solely to satisfy this skill.
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.

