메뉴
Continue working on a story or task by reading progress.md, executing the next work unit, and updating progress. Use when user wants to resume story/task work or mentions /continue command.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Create a new task (bug, refactoring, or qa) with spec and progress tracking. Use when user wants to create a task or mentions /task command.
Run adapter module tests. First argument is adapter name (matches directory under backend/adapters/). Use when user wants to run adapter tests or mentions /test-adapter command.
Run tests with coverage and report uncovered lines/branches per class. Use when user wants to check test coverage, find untested code, identify coverage gaps, or mentions /test-coverage command. Also use after green phases to verify new code is well-covered.
Review tests to replace loose validation (contains, isNotNull, isNotEmpty) with strict validation (isEqualTo on parsed fields). Use when user wants to improve test assertions or mentions /test-review command.
| name | continue |
| description | Continue working on a story or task by reading progress.md, executing the next work unit, and updating progress. Use when user wants to resume story/task work or mentions /continue command. |
ProductSpecification/stories.md, move it to In Progress before proceeding[~] or [ ] entrycarryover.md (story root, if it exists) and the current scenario's summary file (summaries/{scenario-slug}.md, if it exists). Treat both as additional context for the work unit -- they preserve predictions, decisions, surprises, and quirks from prior conversations. /continue only READS these files; it never writes them (the /handoff skill is the sole writer).decisions/*-decision.md files in the story directory. If any exist AND the current step references the ADR (via "see ADR" annotation or matching scenario), read it. ADRs contain architectural decisions, schema changes, edge cases, and implementation guidance that the work unit needs.[ ] adapters-discovery, read usecase constructor to identify ports and map to adapters (see .claude/guidelines/workflow-detail.md). Mark [x] adapters-discovery, insert concrete steps below it, commit progress.md. The bug-task [ ] steps discovery gate resolves the same way via its Work Unit Dispatch row -- run its hazard-catalogue fan-out, then insert the TDD steps.ProductSpecification/stories.md (see below)ProductSpecification/stories.md for stories). If ALL checkboxes are now [x] or [S] (no [ ] or [~] remaining), move the task folder to ProductSpecification/tasks/done/ and include the move in this commit./refactor and the two review passes (agent-review-agent + premortem-agent) concurrently over the behavior commit (see "Pre-Commit Review Passes" below); the passes are non-gating -- collect their verdicts for the report. Then the refactor commit (/refactor's changes only; skipped if it changed nothing). Skip both /refactor and the passes for a progress-only behavior commit. Fold the review verdicts into the stop-and-report.The high-level lifecycle, status markers, and atomic-unit rule are in .claude/rules/workflow.md; the detailed scenario sequences, adapter-discovery procedure, progress mechanics, and task sequences are in .claude/guidelines/workflow-detail.md. Progress file format examples are in .claude/templates/workflow/progress-format.md.
| Argument | Resolution |
|---|---|
task N | Find ProductSpecification/tasks/N-*/progress.md |
| Bare number or name | Resolve story via ProductSpecification/stories.md then ProductSpecification/stories/NN-story-name/progress.md |
| No argument | Scan recent git log for Story N or Task N references; most recent wins |
File lookup: Use find via Bash (not Glob) when searching for progress files or story folders. Glob is unreliable on Windows/MINGW with large .gitignore files. For story resolution, derive the folder name from ProductSpecification/stories.md (e.g., story 5 "Create a task" → 05-create-task) and Read the progress file directly. Use ls or find via Bash only when the folder name is ambiguous.
Each progress.md checkbox maps to sub-skills. Dispatch per .claude/guidelines/workflow-detail.md sequences. This table applies equally to stories AND tasks — never skip /test-review or /refactor for task steps.
| Checkbox | Sub-skills |
|---|---|
Spec items (interview, story, mockups, api-spec, test-spec) | /{item} then commit |
root cause analysis (bug tasks) | Run inline (no subagent): /rca — re-verify prior assumptions, test competing hypotheses, confirm the cause with real data, record findings in spec.md, mark [x] → progress-only commit |
design | /design-preview → user approves (optionally with ADR) or /architecture → commit (if ADR produced) |
red-* (acceptance, usecase, adapter, selenium, frontend, frontend-api) | red-agent.md → /test-review → commit → /refactor → commit |
green-usecase, green-adapter X | green-agent.md → /test-coverage {module} --focus → commit → /refactor → commit |
adapters-discovery | Load .claude/templates/workflow/adapter-discovery-checklist.md, run all 3 checks (ports, exceptions, response shape), mark [x] adapters-discovery, insert concrete red-adapter X / green-adapter X steps (or [S]) → commit progress.md |
steps discovery (bug tasks) | Run the hazard-catalogue fan-out per .claude/guidelines/workflow-detail.md "Hazard scan at steps discovery" (one hazard-scan-agent per group in _index.md, concurrently, then a synthesis pass over the seams); fold each fired-trigger GAP in as a discovered red-* step or dismiss it with a reason (an unresolved GAP blocks insertion); resolve the gate with its REQUIRED scan record [x] steps discovery (scope: <layers>; scanned all _index.md groups; GAPs: <folded as red-* / dismissed: reason / none fired>) — a bare [x] steps discovery with no scanned/GAPs clause is an unscanned gate, not a valid completion; insert the concrete TDD steps below the gate → commit progress.md |
green-acceptance | Run inline (no subagent): read green-agent.md workflow, load acceptance implementation template, enable the disabled test (remove disable marker — only allowed test change), run acceptance tests, verify GREEN → commit |
green-frontend, green-frontend-api | green-agent.md → commit → /refactor → commit |
green-selenium | /run-backend → /run-frontend → green-agent.md (remove-marker-only: no production code, no Statements changes, no backend changes — if test fails, STOP and report) → commit |
align-design | Build component → /align-design → /design-review (MANDATORY) → /test-coverage frontend --focus → commit → /refactor → /align-design verify-only → commit |
demo | /demo {scenario_test_class} then progress-only commit |
refactor usecase / refactor (...) | Apply change then run affected tests then commit |
QA ## Cases checkbox | No dispatch. Report the next unchecked case to the user and stop -- the tester verifies it manually against the target environment, then ticks the box (or files a separate bug task if it fails) on their own. |
A single /continue invocation executes exactly ONE work unit. Within that work unit, don't pause between sub-skills. A /refactor work unit ends in two commits — behavior commit, then separate refactor commit; STOP only after the final commit, NEVER after the behavior commit while /refactor is still pending. The two review passes run just before the work unit's final commit, in the /refactor batch (skipped for progress-only commits). Once the final commit lands, stop and report: completed step, test results (pass/fail counts from every test run in the work unit), the review-pass verdicts, next step, progress fraction, how to continue. Do NOT read the next [ ] step and keep going.
End the report with a one-line /plain hint (e.g. Press /plain to have this report re-explained in plain words) so the user has a button to press when the technical summary didn't land. This is only a pointer -- never invoke /plain yourself; it is a manual button the user presses.
Test results: Collect pass/fail counts from every sub-skill that runs tests (red-agent, green-agent, test-coverage, refactor). Include them in the final report as a summary line, e.g., Tests: 15 passed, 0 failed or Tests: 14 passed, 1 failed. When multiple test suites ran, report each separately.
Red prediction (mandatory for red- work units):* When the work unit included any red-* phase, copy the red-agent's Predicted failure, Actual failure, and Comparison sections verbatim into the final report — same wording as the Output Summary Format in .claude/templates/workflow/red-phase-formats.md. Do NOT collapse to phrases like "test passed as predicted" or "prediction matched" — the user must see both the prediction and the actual result side by side, in their own labelled sections, so the match can be audited without re-reading the agent's return.
Review-pass findings (mandatory when the review passes ran): include the agent-review and premortem verdicts. PASS → one line each (agent-review: PASS, premortem: PASS). CONCERNS/BLOCK → list each finding with its place in the diff and the named missing guard, as a follow-up the user can act on. These never revert the commit; they are surfaced, not auto-fixed.
After the behavior commit lands, and concurrently with /refactor, dispatch two
fresh-context passes over that commit, in one message as two concurrent Agent
calls — so they overlap /refactor instead of adding a serial tail to every work unit:
agent-review-agent — audits what the work contains: any problem, unnarrowed.premortem-agent — imagines what the work is missing: the incident it would cause.Input is the behavior commit, not a working-tree snapshot. The behavior commit is
already an immutable freeze of the reviewable work — the primary change plus the
progress.md/stories.md advances — so pass each pass the behavior commit's range (HEAD
after the behavior commit, or HEAD~N..HEAD if the behavior phase produced N commits) plus
one line of context on what the unit did. Reading a committed range is what makes the
overlap safe: /refactor edits the live tree concurrently, but the passes read the
immutable commit, so there is no read/write race and no snapshot file to manage. Because
they read the behavior commit, the passes do not see /refactor's changes — an accepted
residual: /refactor is behavior-preserving and its own test run gates it green before the
refactor commit, so the only thing left unreviewed is a structural delta that changes no
behavior.
Non-gating. The refactor commit lands regardless of verdict: a CONCERNS or BLOCK
surfaces in the stop-and-report as a follow-up — it does NOT block, revert, or amend either
commit. PASS is silent. Running them in the /refactor batch changes when they read,
never their authority — they stay a surfacing layer, not a gate. The only two gates that
block a commit remain /test-review and /refactor.
Skip the passes only for a progress-only commit — a checkbox flip has nothing to audit.
Run them on any unit that ships reviewable content: code, tests, specs, or prompt
artifacts. When a unit has no /refactor step (e.g. green-acceptance, green-selenium),
it produces a single commit; dispatch the two passes over that commit right after it lands.
Why these run as a separate layer — defense-in-depth that differs in kind from the
in-loop /test-review and /refactor — is in
.claude/guidelines/workflow-detail.md "Pre-Commit Review Passes".
Before the behavior commit, verify: (1) primary skill ran, (2) /test-review ran (red phases), (3) /test-coverage ran (green-usecase/green-adapter). /refactor and the two review passes are not in the behavior commit — they run after, in the /refactor batch, before the final commit. Before the final commit, verify: (4) /refactor ran (all phases except green-acceptance/green-selenium/demo/spec items), (5) the two review passes ran over the behavior commit (every reviewable-content unit; skip only progress-only commits). If /refactor or a review pass was skipped -- run it before the final commit.
ALL sub-skills dispatch via Agent tool for context isolation:
| Sub-skill | Dispatch method |
|---|---|
red-* | Agent tool (subagent_type: red-agent) — pass layer, story folder path, scenario name, and ADR content (if loaded) |
green-* (except green-acceptance) | Agent tool (subagent_type: green-agent) — pass layer, story folder path, scenario name, and ADR content (if loaded) |
green-acceptance | Inline — no subagent. Main agent reads green-agent.md, loads acceptance template, enables the test, runs it. Full visibility for user. |
/refactor | Agent tool (subagent_type: refactor-agent) |
/test-review | Agent tool (subagent_type: test-review-agent) |
/test-coverage | Agent tool (subagent_type: coverage-agent) |
agent-review (pre-commit) | Agent tool (subagent_type: agent-review-agent) — pass the behavior commit's range (HEAD, or HEAD~N..HEAD) + one line of intent. Dispatched in the /refactor batch, concurrently with premortem. |
premortem (pre-commit) | Agent tool (subagent_type: premortem-agent) — pass the behavior commit's range (HEAD, or HEAD~N..HEAD) + one line of intent. Dispatched in the /refactor batch, concurrently with agent-review. |
Derive the layer from the checkbox (e.g., red-adapter storage → layer storage, green-usecase → layer usecase). Both red-agent and green-agent receive: layer, story folder path, scenario name, and ADR content (if loaded in step 5). The agent resolves test files and templates from its own workflow.
CHAINING: After each sub-step completes (Agent tool return), echo a 1-2 line status summary (agent name, outcome, pass/fail counts) to the user, then immediately dispatch the next sub-step. Do NOT wait for user input between sub-steps — the echo is informational only. (The two review passes are the exception: they dispatch together in one message, in the same batch as /refactor.)
AGENT LOG: Before the first agent dispatch, clear the log: > infrastructure/agent-progress.log. The two review passes log too (they run in the /refactor batch, just before the final commit); after the final commit, include the log contents in the stop-and-report summary.
LOG REMINDER: Every time you dispatch a sub-agent (Agent tool call), output this line immediately before the call:
> Dispatching {agent-name}. Live progress: tail -f infrastructure/agent-progress.log
This reminds the user where to watch. The line appears in conversation output before the agent starts, so the user can open a terminal and tail the log while the agent works.
/refactor batch). Never stop between sub-skills./refactor work unit ends in two commits: the behavior commit (carries the progress.md advance), then a separate refactor commit (skipped if /refactor changed nothing). Otherwise one commit carries progress.md.task: (e.g., task: red-adapter storage (Task 1, Step 1)).claude/guidelines/workflow-detail.md sequencesAfter updating progress.md for a story (not tasks), update the story's row in ProductSpecification/stories.md to reflect current phase status. The file has columns: Spec | Backend | Integration | Frontend | Security | Load | Infra | Status.
Phase column values — derive from progress.md sections:
✅ — all checkboxes in that section are [x] or [S]🔧 — section has at least one [~] or [ ] checkbox (work in progress or next up within the current active lifecycle phase)— — section exists but no work started yet (all [ ])n/a — phase not applicable (test spec file says "No tests" or story has no scenarios for that phase)· — no story folder or no progress file existsTests and % columns — recount after every progress.md update:
### scenario headings in progress.md (not ## section headers)[x] or [S] (no [ ] or [~] remaining)round(done / total * 100)done/total in Tests column, N% in % columnStory completion — when all scenarios reach 100% (Tests column = total/total, % = 100%), move the story row from the In Progress table to the Done table in ProductSpecification/stories.md. Keep all column values intact.
When to update: after every progress.md commit for a story. Include ProductSpecification/stories.md in the same commit.
Lifecycle ordering: Spec → Backend → Integration → Frontend → Security → Load → Infra. A phase becomes 🔧 when it is the current active phase (has [~] items) OR when it still has [ ] items and all prior phases are ✅.
.claude/templates/workflow/progress-format.md -- progress file format for stories, bug tasks, and refactoring tasks