Skip to content
Back to skills

Analyze

ASecurity

Cross-check spec, plan, and tasks for consistency. Phase 4.5 of the eight-phase workflow. Catches scope drift, missing acceptance criteria, dependency cycles. Modeled on github/spec-kit /speckit.analyze. Run after /tasks, before /implement.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 9, 2026
ai-agentsgit

Works with

  • cli
  • mcp

Security analysis

A100/100

Scanned October 9, 2026

npx -y skills add qyndex/claude-init --skill analyze --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Analyze?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Analyze
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/qyndex-analyze/badge)](https://www.skillsdirectory.com/skills/qyndex-analyze)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: analyze
description: Cross-check spec, plan, and tasks for consistency. Phase 4.5 of the eight-phase workflow. Catches scope drift, missing acceptance criteria, dependency cycles. Modeled on github/spec-kit /speckit.analyze. Run after /tasks, before /implement.
when_to_use: Spec, plan, and tasks all exist for a feature and you want a sanity check before coding. User says "analyze the plan", "check consistency", "any gaps".
argument-hint: "<feature id>"
model: sonnet
allowed-tools: Read, Glob, Grep, TodoWrite, Write
---

# Analyze

Cross-artifact consistency check. Catch drift before it becomes code.

## Process

1. Read the **spec** (`specs/active/<id>-<slug>.md`).
2. Read the **plan** (`plans/active/<id>-<slug>.md`).
3. Read the **tasks** in `tasks/TASKS.md` filtered by `spec:<id>`.
4. Run the consistency checklist (below).
5. Report findings, severity-tagged.
6. **Write the machine marker** (e2e-audit spec-pipeline-1 — this is what makes the
   gate real; without it next-task.sh/workflow-state.sh see the spec as unanalyzed):
   Write `.claude/state/analyze-<id>.json` with exactly:
   ```json
   {"verdict": "PASS|BLOCK", "blockers": ["<each BLOCKER finding>"], "date": "<ISO>"}
   ```
   `<id>` is the numeric spec id (e.g. `analyze-042.json`). Re-running /analyze
   overwrites the marker — verdict reflects the LATEST run.
7. Block `/implement` if any blockers — the marker's `verdict: BLOCK` is enforced
   by `next-task.sh --require-analyze` (loop path), not just by this prose.

## Consistency checks

- [ ] Every spec acceptance criterion has at least one task that exercises it.
- [ ] Every plan component is implemented by at least one task.
- [ ] No task references a file outside the plan's scope.
- [ ] No cyclic dependencies in the task DAG.
- [ ] Every task's `accept:` command is syntactically valid (no obvious typos).
- [ ] All declared dependencies (packages, services, MCP servers) appear in install/setup tasks.
- [ ] Spec, plan, and tasks share the same id namespace.
- [ ] No open `[OQ]` items in the spec.
- [ ] Phase boundaries align — every plan phase has tasks; no orphan tasks.
- [ ] Risks in the plan have mitigation tasks or explicit acceptance of the risk.
- [ ] Backwards-compat constraints from the spec appear as tests.

## Output

```
# Analysis — feature <id> — <date>

## Coverage
- Spec criteria with tasks: 8/8 ✓
- Plan components with tasks: 6/6 ✓
- Tasks with accept commands: 12/12 ✓

## Gaps
- BLOCKER: Acceptance criterion "rate limit 100 req/min" has no task.
- HIGH: Plan mentions `redis-rate-limiter` but no install task.

## Cycles
- None ✓

## Verdict
BLOCK — 1 blocker, 1 high. Address before /implement.
```

After save: post the summary to the user. If green, "Ready for /implement." If red, list the gaps.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.