| name | spec-orch-spec-stage |
| description | Drive the SpecOrch Spec stage from a coding environment (Cursor/Claude Code). Use when the user wants to plan a feature, draft a spec, manage questions, approve a spec, or advance an issue through the SpecOrch lifecycle. Triggers on "spec-orch", "plan feature", "draft spec", "questions", "approve spec", "advance issue", or "EODF workflow". |
SpecOrch Spec Stage โ Coding Environment Driver
This skill replaces the autonomous LLM Planner when working inside a coding
environment. You (the AI assistant) act as the planner โ analysing the issue,
generating questions, drafting the spec, and driving the state machine via CLI.
Lifecycle States
DRAFT โ SPEC_DRAFTING โ SPEC_APPROVED โ BUILDING โ VERIFYING โ REVIEW โ GATE โ ACCEPTED
Workflow
1. Create or load the issue fixture
spec-orch plan-to-spec docs/plans/my-feature.md --issue-id SPC-FEAT-1
2. Draft the spec
spec-orch spec draft SPC-FEAT-1
This creates an initial spec_snapshot.json in the workspace.
3. Add questions (your analysis)
Analyse the issue and add clarifying questions:
spec-orch questions add SPC-FEAT-1 \
--text "Should the API support pagination?" \
--category requirement \
--blocking
spec-orch questions add SPC-FEAT-1 \
--text "Which database backend?" \
--category architecture \
--blocking
Categories: requirement, environment, architecture, risk.
4. List and answer questions
spec-orch questions list SPC-FEAT-1
When the user provides answers:
spec-orch questions answer SPC-FEAT-1 q-abc12345 \
--answer "Yes, cursor-based pagination" \
--decided-by chris
5. Approve the spec
Once all blocking questions are answered:
spec-orch spec approve SPC-FEAT-1 --approved-by chris
Check the spec status anytime:
spec-orch spec show SPC-FEAT-1
6. Advance to build
spec-orch advance SPC-FEAT-1 --live
Or run directly:
spec-orch run-issue SPC-FEAT-1 --live
7. Post-build workflow
spec-orch review-issue SPC-FEAT-1 --verdict pass --reviewed-by chris
spec-orch accept-issue SPC-FEAT-1 --accepted-by chris
spec-orch findings list SPC-FEAT-1
If verification fails or you fixed code after review feedback, re-run checks:
spec-orch rerun SPC-FEAT-1
Decision Guide
| Situation | Action |
|---|
| New feature request | plan-to-spec โ spec draft โ add questions |
| User answers a question | questions answer |
| All questions answered | spec approve |
| Spec approved | advance or run-issue |
| Build complete | review-issue โ accept-issue |
| Need to re-verify | rerun |
See also (complementary)
- MR โ Review โ CI โ Merge โ Linear๏ผไปๅบๅ
.cursor/skills/spec-orch-mr-closed-loop/SKILL.mdใๆฌ Skill ๅช็ฎก Spec/้ฎ้ข/ๆนๅไธ CLI ็ถๆๆบ๏ผๅๅ
ฅ main ็ไบคไป่ๅฅ็ฑ่ฏฅ Skill ็บฆๆใ
Key Principle
The coding environment AI is the planner. There is no need for LiteLLM or any
external LLM call. You analyse the issue using your own intelligence, generate
questions via CLI, and drive the state machine step by step. The same CLI
commands also work in Daemon mode with an autonomous LLM planner โ the
artifacts and state transitions are identical.