spec-validate
Skill1 starsChanged 5 months agoMITView on GitHub
Use when auditing spec completeness before merging a feature branch, when broad spec gaps are suspected, or when referred from spec:maintain or spec:core.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- renanlido/tia-harnesstheir own version, first seen Nov 1, 2026
- simonbbyrne/claude-skillstheir own version, first seen Mar 5, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- renanlido/tia-harness★ 12same name
- simonbbyrne/claude-skills★ 1same 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 repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- spec-bootstrapSkill
- spec-captureSkill
- spec-coreSkill
- spec-inferSkill
- spec-maintainSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 20
Check every category, in order.
- Distinctive: line 109
If a category has no findings, omit it from the report.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Spec: Validate
Overview
Validate audits the full specs/ directory against the codebase and produces a structured gap report. It runs without clarification — audit first, questions after. Do not ask the user anything before producing the report.
Severity Levels
- Critical — rebuild from specs would produce wrong behaviour
- Gap — rebuild would produce incomplete behaviour
- Drift — spec and code have diverged but rebuild intent is recoverable
The Ten Categories
Check every category, in order. Do not skip any. Report findings under the appropriate severity.
1. Missing capability specs
Code areas with no corresponding spec file, or not listed in specs/README.md.
How to find: scan src/ (or equivalent) for functional areas; cross-reference with specs/.
Severity: Critical if a core area; Gap if peripheral.
Referral: spec:infer
2. Outdated decisions
Spec entries contradicted by current code.
How to find: for each Decision entry, verify the described behaviour exists in code.
Severity: Critical if the contradiction would produce wrong rebuilt behaviour; Drift if intent is recoverable.
Referral: spec:maintain if from a recent change; spec:infer if drift is older or scope is unclear.
3. Unconfirmed inferences
Remaining > Inferred — please confirm markers in any spec file.
How to find: grep specs/ for > Inferred.
Severity: Gap — the entry has not been validated by a human.
Referral: review directly with the user; spec:infer as complement if code provides more context.
4. Missing invariants
Capability specs that have a ## Decisions section but no ## Invariants section.
Severity: Gap.
Referral: none (spec gap, not a skill referral — add invariants when updating the spec).
5. Untested invariants
Invariants with no corresponding test in the test suite.
How to find: for each INV-NN: entry, search the test suite for coverage of that assertion.
Severity: Gap — spec is complete but rebuild guarantee is unverifiable.
Referral: none (testing gap, not a spec gap — surface as finding only).
6. Incomplete ADRs
ADR files missing required sections: Status, Context, Decision, Rejected alternatives. Severity: Gap. Referral: none.
7. Superseded ADRs still being followed
Code that follows a superseded ADR rather than the latest accepted one.
How to find: check each ADR marked Superseded by ADR-NNN; verify code follows the superseding ADR.
Severity: Critical if the superseded decision produces wrong behaviour; Drift if recoverable.
Referral: spec:maintain.
8. CLAUDE.md enforcement block missing or incomplete
CLAUDE.md (and AGENTS.md) must contain the spec-driven development section with the full skill list.
Severity: Gap.
Referral: spec:bootstrap (adds the block if never created).
9. specs/README.md out of sync
Files in specs/ not listed in specs/README.md, or entries in specs/README.md with no corresponding file.
Severity: Drift.
Referral: none — update specs/README.md directly.
10. Unpromoted or fragmented cross-cutting ADR clusters
Branch-level audit. Per-commit spec:maintain invocations are structurally blind to cross-cutting clusters that emerge across commits, and may also fragment one cluster into multiple per-commit ADRs. This category catches both.
How to find:
- Walk commits since branch-from-main (
git log main..HEAD). - For each commit's capability spec edits, note: specs touched, presence of rejected alternatives, presence of directional-shift language ("from X to Y", "no longer", "moving from", "shifted to").
- Identify candidate ADR clusters: the aggregate set of capability specs touched across the branch is 2+ (or a framework/principle is introduced that lives in no single capability), AND a shared directional argument or a rejected-alternative cluster spans the commits.
- For each candidate cluster:
- No ADR exists for it → missing ADR finding. The cluster needs one consolidated ADR.
- Multiple per-commit ADRs each capture a fragment of one cluster → fragmented ADR cluster finding. Recommend consolidation into one ADR.
- A stand-alone ADR cleanly captures the cluster per the
spec:core(a) + (b or c) heuristic → clean.
Severity: Gap (missing ADR); Drift (fragmented cluster — consolidation recoverable).
Referral: spec:maintain to author the consolidated ADR; spec:capture to assemble the cross-cutting view from session history when available.
Report Format
## Spec Validation Report
### Critical
[ID] [file] — [description]
→ [referral skill if applicable]
### Gap
[ID] [file] — [description]
→ [referral skill if applicable]
### Drift
[ID] [file] — [description]
→ [referral skill if applicable]
### Summary
X critical, Y gaps, Z drift items.
Recommended next step: [highest-priority action with skill referral]If a category has no findings, omit it from the report. If all categories are clean, say so explicitly — do not silently produce an empty report.
Soft Referrals Per Finding
| Finding type | Referral |
| Missing or outdated capability specs | spec:infer |
| Unconfirmed inferences | Review with user; spec:infer as complement |
| Gaps from a recent change | spec:maintain |
| Broad gaps across multiple areas | spec:infer scoped by area |
| Missing tests for invariants | Surface as finding — no skill referral (testing gap) |
| Missing CLAUDE.md block | spec:bootstrap |
| Missing or fragmented cross-cutting ADR cluster | spec:maintain (to author the consolidated ADR); spec:capture (if session context can supply the cross-cutting view) |
Common Mistakes
| Mistake | Correct behaviour |
| Asking clarifying questions before running | Audit mode — produce report first, questions after |
| Stopping after finding the first issue | Check all ten categories, every time |
| Reporting without skill referrals | Every finding gets a referral where one applies |
| Silently skipping categories with no findings | Report clean categories as clean, or omit cleanly — never silently skip |
| Mixing severity levels inconsistently | Critical = wrong rebuild; Gap = incomplete rebuild; Drift = recoverable divergence |
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.

