zerospec

name	zerospec
description	Use only for ZeroSpec documentation workflows: init-scan/init-build AGENTS.md, update or audit AGENTS.md, check SPEC-vs-code drift, or create ZeroSpec SPEC / ADR / SA documents. Do NOT use for application code changes, tests, debugging, generic specs, general architecture discussion, or generic "update" tasks unless the request explicitly mentions ZeroSpec, AGENTS.md, or ZeroSpec-managed docs.

ZeroSpec Skill Router

This skill routes ZeroSpec documentation tasks to the correct prompt pack and applies a built-in Self-Review Protocol after task completion.

---

Step 1 — Identify Task

Match the user's request to one row in the Route Table:

Task	Trigger Keywords	Prompt File
Scan a project before creating AGENTS.md	zerospec init-scan, scan project for AGENTS.md, analyze repo for ZeroSpec	prompts/INIT-SCAN.md
Build AGENTS.md for a new project	zerospec init-build, create AGENTS.md, initialize AGENTS.md	prompts/INIT-BUILD.md
Update an existing AGENTS.md	update AGENTS.md, sync ZeroSpec docs, ZeroSpec docs out of date, zerospec update	prompts/UPDATE.md
Evaluate AGENTS.md quality	audit AGENTS.md, zerospec audit, score AGENTS.md, grade ZeroSpec docs	prompts/AUDIT.md
Check SPEC vs code drift	ZeroSpec SPEC drift, SPEC vs code, spec stale, zerospec drift	prompts/DRIFT.md
Write a SPEC document	write ZeroSpec SPEC, zerospec spec, ZeroSpec API interface contract	prompts/SPEC.md
Write an ADR	write ZeroSpec ADR, ZeroSpec architecture decision record	prompts/ADR.md
Write a System Analysis document	write ZeroSpec SA, ZeroSpec system analysis document	prompts/SA.md

If the request does not match any row, ask: "Which ZeroSpec task do you want to run?" and list the 8 options above.

---

Step 2 — Execute Prompt

1. Read the matched prompt file (linked above) in full.
2. Follow its instructions against the user's project exactly as written.
3. Cover every relevant required section. Mark an inapplicable section with a one-line reason, and do not add unrequested content.

---

Step 3 — Self-Review Protocol

After completing the task, silently apply the matching row below. Fix any issues before outputting the final result — do not show the checklist to the user.

Task	Self-Review Checklist
init-scan	① Scan findings match the actual filesystem ② No key directories or config files missed ③ Greenfield/Brownfield judgment backed by concrete evidence
init-build	① Output follows ZeroSpec structure (Quick Constraints / Domain Map / Commands) ② No fabricated file paths or commands ③ Every INIT-SCAN finding is addressed
update	① Changes reflect actual code evolution ② No contradictions with existing doc content ③ No paragraph bloat (old content trimmed when new content added)
audit	① Score totals match dimension-by-dimension analysis ② Every Fix item references a dimension ③ No over- or under-scoring
drift	① Every DRIFTED verdict cites a specific file path or line ② CLEAN verdicts have no overlooked inconsistencies ③ Severity levels are justified
spec	① All described endpoints/behaviors are covered ② No fabricated request or response fields ③ Changelog format consistent with existing SPECs
adr	① Options pros/cons are unbiased and complete ② No missing alternatives ③ Conclusion follows logically from the analysis
sa	① System boundary matches actual architecture ② No missing module dependencies ③ No implementation details (architecture level only)