goal.md

description	Execute ALL tasks in order until done (goal mode)
argument-hint	[optional limit, e.g.: 3 tasks] [--light]

Read .pi/skills/fia/SKILL.md and the cookbook .pi/skills/fia/cookbooks/harness_bridge.md, and follow Step 3 (goal mode) to the letter.

Before any task: read ai-docs/stack.md. A "decide later" layer that any task will touch = STOP and resolve it with me first (cookbook stack.md, same flow as /stack) — implementing with an undecided stack is guaranteed rework.

Limit/instructions from the engineer: $@

Token discipline — your own context is a cost. This session pays for every line it carries forward (a real project's goal session burned 240M tokens mostly re-reading its own history). Never paste briefs, reports or FDA output into the conversation — reference paths; the viewer (npm run fda:viewer) is the log. Report each task in ONE line. Never re-read task-master.md in full — the sequencer owns it. For routine milestones prefer --light (fda_plan_build_test); the full fda_sdlc is for bigger/riskier tasks.

FDA selection (evaluate after each brief is written): an exact Mode: prototype line selects fda_prototype.mjs for that brief. Otherwise, if the engineer's instructions contain --light, use fda_plan_build_test.mjs; the default is fda_sdlc.mjs. Prototype takes precedence over --light because it is the brief's explicit per-task opt-in.

Loop, until no unblocked task remains OR a milestone boundary closes (step 3):

1. task-sequencer → next unblocked issue → brief in ai-docs/actual-todo/ (sequencer stopped on the theme gate — right after the greenfield foundation task, no closed theme decision log? PAUSE the loop and resolve with me per the cookbook: /theme (recommended) or record my explicit acceptance of the default theme. Never decide for me, never skip silently. Stopped on the env gate — foundation task picked but node imp/scripts/env-preflight.mjs reports dev keys missing from .env.local? PAUSE and provision with me mid-goal per the cookbook: CLI parts yourself, dashboard keys from me — then continue. Stopped on an impossible/circular dependency with a recommended split (Task 06 needs Task 07's schema, 07 is blocked by 06)? Apply the split once — do not ask — and re-delegate. If the second pass still cannot write a brief, THEN ask.)

2. node imp/<fda>.mjs ai-docs/actual-todo/<brief>.md — ONE task per run, never batch them (<fda> = fda_prototype for Mode: prototype; otherwise fda_plan_build_test when --light, fda_sdlc by default)

3. exit 0 → report to me in one line (task, phases, tokens, commit) and continue — then inspect ai-docs/milestones.md plus task statuses. If that task made every task of a milestone done, run the milestone QA immediately, before selecting any task from a later milestone:

   node imp/fda_qa.mjs "<milestone-id>"

   A UI milestone requires a valid ai-docs/ui/contract.json; the QA FDA resolves named rules to APPLY or reasoned SKIP; only browser/audit evidence earns PASS. It blocks on missing or invalid contracts, failed Playwright evidence, or a failed design audit. An API-only milestone exits 0 with a durable Status: skipped report and its reason. QA exit 0 → report the QA report path, then drain the milestone's documentation in ONE call (per-task runs skip their document phase by default — sdlc.document: per_milestone):

   node imp/fda_document.mjs "Document milestone <milestone-id>: what shipped across its tasks — read ai-docs/milestones.md and the completed briefs in ai-docs/todos/, write/update the docs, and commit them"

   — then, when other milestones remain, END THIS LOOP at the milestone boundary: print a 5-line handoff (milestone done, QA report path, next milestone id, "state is durable in ai-docs/ — run /goal again in a NEW session to continue"). Every loop artifact lives on disk, so the fresh session loses nothing and starts without this conversation's accumulated weight — one milestone per session is the cheap shape. Only when NO unblocked task remains anywhere do you continue to the final delivery below. QA exit != 0 → STOP the goal; show the phase, rule id/evidence, and artifact path. Never silently defer QA, continue into the next milestone, or mark a failed milestone complete. This automatic boundary supersedes the cookbook's legacy “suggest /qa only” sentence.

4. exit != 0 → recover automatically while the run is making PROGRESS (cookbook harness_bridge, "On failure"): each failure that names a NEW gap an FDA can apply is yours to repair without asking (--fda-id … --resume, with a verdict --missing naming the gap). A plain re-run is legitimate ONLY when the tree changed since the failure or an engine died — the runner now REFUSES a bare resume over an identical tree, and counts every bare resume against the run's combined recovery budget. Treat any such refusal as STOP-and-ask, never work around it (--retry-unchanged is the engineer's override, not yours). Report each recovery to me in one line (what failed → what you fixed) and continue. Fixing one gate violation often reveals the next; that chain is progress, walk it. From the THIRD recovery of the same run onward, prefer converting the remaining gaps into a FOLLOW-UP brief via the task-sequencer — a fresh run starts with small sessions and clean context, resuming a long run re-opens its giant sessions at full width. STOP only when: the SAME violation comes back (no progress), the outcome is no_progress / attempt_cap / budget_exhausted / engine_exhausted, or verdict.mjs set / the resume itself refuses because the run's recovery budget is spent (both caps are enforced in code — never work around them). On STOP: show the phase, the gate violations, the trace AND your recommended fix. I decide: fix, skip or re-run — and if I answer "continue" (or equivalent), that IS the authorization for exactly the fix you recommended: execute it, never re-ask.

Briefs with a Spec: NNNN (…) line arm the FDA's spec-coverage gate — on success, check the spec's Traceability table reflects the new tests and flag it if it doesn't. The tested FDAs then run spec_close in code: when the current task is the last linked task, it appends the Delivery Gate and sets Status: done before committing. A spec_close pending result is actionable planning metadata in the trace, never a transition to remember after the loop.

Suggest I keep npm run fda:viewer open in another terminal to watch live.

When everything is done, fulfill Step 4 of the cookbook (deliver it RUNNING) before the summary: the app starts with ONE command (npm run dev — if it needs 2+ processes, dispatch an FDA to create the single script), env/database ready and smoke-check done. Whatever requires human action (login, account), ask for it in the MIDDLE of the goal and continue — no homework at the end. Final summary: tasks, commits, tokens — and ALWAYS closing with the How to test section (minimal command, URL, short checklist of what was delivered) + an offer to start the dev server yourself right now.

Before declaring the goal complete, verify every completed UI milestone has a passing report in ai-docs/qa/ (and every API-only milestone has its reasoned skipped report). A missing report is a blocking unfinished boundary: run node imp/fda_qa.mjs "<milestone-id>"; failure means STOP.

Running locally and tested? The next rung is /launch — suggest it at the end (it puts the app on a public URL and then into real production, guided).