Files

Tom Boucher 1e6737cd8e feat(plan-phase): --research-phase flag + scrub stale slash-command refs (#3042 , #3044 ) (#3045 )

* feat(plan-phase): --research-phase flag absorbs deleted /gsd-research-phase + scrub stale refs (#3042, #3044)

#3042 (orphaned research-phase): /gsd-research-phase had a workflow file
but no slash-command stub. Rather than restore the orphan, the research-
only capability is now a flag on /gsd-plan-phase:

  /gsd-plan-phase --research-phase <N>

When set, the workflow scopes to phase N, runs the research step (Section
5 of the existing plan-phase workflow), then early-exits before the
planner/plan-checker/verifier chain.

Per RCA against the deleted standalone, the flag adds two modifiers to
fully cover the original surface (Option B from the RCA discussion):

- --view : print existing RESEARCH.md to stdout, no spawn. Cheapest mode
  for the correction-without-replanning loop the issue reporter
  explicitly called out. Errors with a clear hint if RESEARCH.md is
  missing.
- --research : reuse the existing "force re-research" semantics. In
  research-only mode this skips the existing-RESEARCH.md prompt and
  re-spawns unconditionally.
- Neither flag, RESEARCH.md exists : prompt update/view/skip. Mirrors
  the deleted standalone's existing-artifact menu (#3042 RCA).

#3044 (stale slash-command refs): scrubbed five deleted commands from
all user-facing surfaces, including English docs, 4 localized doc sets
(ja-JP, ko-KR, zh-CN, pt-BR), workflows, templates, and references.

  /gsd-check-todos          → /gsd-capture --list
  /gsd-new-workspace        → /gsd-workspace --new
  /gsd-status               → /gsd-progress
  /gsd-plan-milestone-gaps  → table rows / orphan sections removed
                              (PR #3038 only scrubbed workflows/agent;
                              missed the docs surfaces this PR covers)
  /gsd-research-phase       → /gsd-plan-phase --research-phase

Includes a fix to docs/issue-driven-orchestration.md (PR #3036)
which itself referenced /gsd-new-workspace 4 times — self-correction.

Removed:
- get-shit-done/workflows/research-phase.md (orphan, capability
  absorbed into --research-phase flag)

Tests:
- tests/bug-3042-3044-research-flag-and-stale-refs.test.cjs — 46
  structural-IR tests across both bugs:
  - argument-hint advertises --research-phase + --view
  - workflow parses --research-phase, sets RESEARCH_ONLY,
    early-exits before planner
  - --view prints RESEARCH.md without spawning
  - --research forces refresh in research-only mode
  - existing-RESEARCH.md prompt path with update/view/skip
  - workflows/research-phase.md is removed
  - 5 deleted slash-commands absent from 17 English user-facing
    surfaces + 16 localized doc surfaces (4 locales × 4 docs each)
  - replacement command tokens present where deleted ones lived

6950/6950 full suite pass. Lints clean.

Closes #3042
Closes #3044

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* fix: address all 8 CR findings on PR #3045

Major (3):
- get-shit-done/workflows/plan-phase.md:344 — added explicit early-exit
  guard at Section 5.1: "Skip if RESEARCH_ONLY=true". Without it, an LLM
  could fall through "use existing, skip to step 6" → planner spawn,
  violating the research-only contract. The guard makes the early-exit
  unreachable from any non-research-only branch.
- get-shit-done/references/continuation-format.md (3 examples) +
  zh-CN/.../continuation-format.md (3 examples) — pointed to
  `/gsd-plan-phase --research-phase` but docs/COMMANDS.md didn't
  document the flag. Added a full --research-phase + --view + --research
  modifier section to the /gsd-plan-phase flag table in COMMANDS.md so
  the canonical reference matches the continuation examples.

Minor (5):
- docs/FEATURES.md:1632 — `/gsd-plan-phase --research-phase` →
  `/gsd-plan-phase --research-phase <N>` (include required arg).
- get-shit-done/templates/README.md:46 — NN-VALIDATION.md producer
  reverted from `/gsd-plan-phase --research-phase` (Nyquist) to plain
  `/gsd-plan-phase` (Nyquist). VALIDATION.md is created during normal
  Nyquist flow, not research-only mode — the bulk replacement was
  wrong for that line.
- get-shit-done/workflows/help.md:89 — signature line was missing
  `--research`; added it alongside `--research-phase` and `--view`.
- tests/bug-3042-3044-...:197 — promptHasView/promptHasSkip were
  tautological (matched anywhere in 1700-line workflow). Tightened
  to a proximity check anchored on "RESEARCH.md already exists" prompt
  header within a 600-char window. Updated workflow to emit that
  literal phrase.
- tests/feat-2840-...:95 — workspace assertion used `/gsd-workspace`
  but the documented replacement is `/gsd-workspace --new`. Tightened
  to require both tokens (in 3 places: requiredCommands list, regex
  in conceptPairs, error message).

6950/6950 full suite pass. Lint clean.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

2026-05-02 23:12:50 -04:00

9.3 KiB

Raw Blame History

Issue-Driven Orchestration with GSD

Status: stable workflow guide Audience: developers who track work in GitHub Issues, Linear, Jira, or similar issue trackers and want to drive AI-assisted implementation through GSD's existing primitives.

What this guide is

A recipe for combining commands GSD already ships into an issue-tracker → workspace → plan/execute → verify/review → PR loop. It is documentation only. No new commands, no daemon, no tracker integration — every command referenced below already exists in GSD today.

The shape is inspired by OpenAI's open-source Symphony orchestration reference (repository). GSD does not vendor or wrap Symphony. The orchestration concepts map cleanly onto primitives GSD already exposes; this guide just spells the mapping out so you can adopt the pattern without writing glue code or bypassing GSD's safety gates.

Why this exists

GSD has the building blocks for issue-driven AI development — /gsd-workspace --new, /gsd-manager, /gsd-autonomous, /gsd-verify-work, /gsd-review, /gsd-ship, plus STATE.md and the phase artifact suite — but no guide that walks through how to drive them from a single tracker issue without writing custom orchestration scripts. Without that guide the failure modes are:

Underuse: developers run discuss/plan/execute manually and never reach for /gsd-manager or /gsd-autonomous even when their work pattern fits.
Workaround scripts: developers wire ad-hoc shell loops between their tracker and claude invocations, bypassing STATE.md, the phase manifest, and the verification gates.

This guide makes the canonical loop discoverable.

Concept mapping

Each row maps a Symphony-style orchestration concept to the GSD primitive that already serves it. Use this table as a translation key when reading Symphony docs, blog posts, or third-party orchestration write-ups.

Symphony concept	GSD primitive
`WORKFLOW.md` (top-level intent)	`ROADMAP.md` (project intent), `STATE.md` (live status), phase `CONTEXT.md` (per-phase scope), phase `PLAN.md` (executable steps)
One isolated agent workspace per task	`/gsd-workspace --new --strategy worktree`
Agent dispatch and concurrency	`/gsd-manager` (interactive dashboard), `/gsd-autonomous` (unattended)
Per-phase plan and discuss steps	`/gsd-discuss-phase` → `/gsd-plan-phase` → `/gsd-execute-phase`
Proof-of-work / test evidence	`/gsd-verify-work` (UAT.md persisted across `/clear`)
Adversarial review	`/gsd-review` (cross-AI peer review of plans)
Human merge gate	`/gsd-ship` (creates PR, optional code review, prepares merge)
Follow-up capture	`/gsd-note`, `/gsd-plant-seed`, `/gsd-new-milestone`, or a manually opened tracker issue
Concurrency control	Manager / background-agent semantics (no always-on poller)

The mapping is one-way: GSD owns the safety gates (verification, human review, explicit confirmation for follow-up creation). Symphony's "continuous orchestration" framing is intentionally not adopted — see Non-goals.

End-to-end flow

The canonical issue → PR loop, written so it can run from a single tracker issue end-to-end. Replace bracketed placeholders before running.

Pick the tracker issue. Choose one issue from your tracker (GitHub, Linear, etc.) that is well-scoped enough for autonomous implementation — bounded scope, observable acceptance criteria, no upstream dependencies that block execution.
Map to a GSD phase. If the issue maps onto an existing phase in ROADMAP.md, select it. If not, run /gsd-new-milestone (for a new milestone of related issues) or open a phase via /gsd-add-phase / /gsd-insert-phase. Capture the tracker issue URL in the phase's CONTEXT.md so traceability survives compaction.
Create an isolated workspace. Run /gsd-workspace --new --strategy worktree <slug> to spin up a git worktree with an independent .planning/ directory. The worktree is the safety boundary: any exploration, partial commits, or aborted plans stay outside main.
Run discuss → plan → execute through GSD. From inside the workspace, run /gsd-discuss-phase to clarify ambiguities, /gsd-plan-phase to produce PLAN.md, and either /gsd-manager (interactive dashboard) or /gsd-execute-phase / /gsd-autonomous (unattended) to implement. Avoid driving raw claude invocations from outside GSD — that bypasses STATE.md updates and the phase manifest.
Demand proof-of-work. Run /gsd-verify-work to walk the user through UAT against the phase's acceptance criteria. Tests, screenshots, log captures, and config diffs are all recorded in UAT.md, which persists across /clear and feeds gaps into /gsd-plan-phase --gaps when verification surfaces missed scope.
Pass through the review and ship gates. Run /gsd-review to get adversarial peer review of the plan from independent AI CLIs (catches blind spots model-by-model), then /gsd-ship to open the PR with a rich body assembled from the planning artifacts. Both gates require a human decision before anything reaches the remote.
Capture follow-up work explicitly. Use /gsd-note for inline notes, /gsd-plant-seed for ideas worth a future phase, or /gsd-new-milestone for a coherent group of follow-ups. Creating a tracker issue from a discovered follow-up requires explicit user confirmation — GSD does not post to remote trackers automatically.

When the PR merges, the loop closes. Auto-close keywords in the PR body (Closes #NNN / Fixes #NNN) close the tracker issue at merge time.

Safety boundaries

The loop is safe because four invariants hold by construction:

Isolated worktrees. Every issue runs in a /gsd-workspace --new worktree, so partial work, aborted plans, and exploratory commits never touch main. gsd-local-patches/ is the recovery surface if a worktree's hand-edits need to come back across an update.
Explicit human review. /gsd-review and /gsd-ship both stop for human approval. There is no auto-merge and no auto-PR-from-execution path. If you want to remove the human gate for a specific repository, that is your branch-protection / merge-queue policy decision, not something GSD opts into for you.
No automatic public posting. GSD never opens, comments on, or closes a tracker issue without an explicit user-initiated command. Follow-up capture defaults to local artifacts (notes, seeds, milestones); pushing back to the tracker is a separate manual step.
Verification before ship. /gsd-verify-work's UAT.md must record evidence before /gsd-ship is run. The recommended discipline is to treat verification_failed as a blocker even when the implementation looks correct — the failure usually surfaces a missed acceptance criterion, not a flaky test.

If any of these invariants is bypassed (e.g. running claude directly against the worktree, skipping /gsd-verify-work, or scripting issue creation through the tracker API without user confirmation), the guarantees of this guide do not apply.

Non-goals

This guide deliberately does not propose any of the following. They are listed here so future contributors don't re-litigate them in code review:

No vendoring or copying Symphony code. GSD reuses its own primitives. The mapping above is conceptual; no Symphony-derived source ships in this repo.
No long-running daemon. GSD does not poll GitHub or Linear. The manager and autonomous workflows handle concurrency through background-agent semantics, not a daemon.
No mandatory tracker dependency. The loop works without any tracker integration. The "tracker issue" step is a human input — the URL goes into CONTEXT.md. GSD has no opinion about which tracker you use, or whether you use one at all.
No bypass of verification, review, or human decision gates. Even when running /gsd-autonomous, the verification and review gates still fire. The "autonomous" label refers to phase-to-phase progression, not to skipping human approval.
No expansion of the default skill / command surface. Every command referenced in this guide already exists. This guide is a documentation surface, not a feature surface.

Possible future follow-up

If maintainer experience with this loop justifies it, a separate approved-enhancement could later add a minimal tracker bridge:

Importing one GitHub or Linear issue into a GSD workspace / phase.
Exporting UAT.md evidence as a comment on the source issue.
Generating follow-up tracker issues from /gsd-plant-seed output.

Each of those would be its own enhancement proposal because each adds integration surface and ongoing maintenance burden. They are out of scope for this guide.

docs/USER-GUIDE.md — task-oriented walkthroughs of individual commands referenced above.
docs/COMMANDS.md — full reference for /gsd-* commands.
docs/FEATURES.md — feature-level capability matrix (workspaces, manager, autonomous, verify, review, ship).
docs/ARCHITECTURE.md — phase-artifact lifecycle and STATE.md mechanics.

9.3 KiB Raw Blame History