FE-864: Brownfield feature delivery from spec — detect, classify, probe, oracle, promote, serve#212
Merged
Merged
Conversation
Contributor
Author
This stack of pull requests is managed by Graphite. Learn more about stacking. |
kostandinang
force-pushed
the
ka/fe-864-orchestrator-enhancements
branch
from
f781af4 to
c2681db
Compare
kostandinang
force-pushed
the
ka/fe-864-orchestrator-enhancements
branch
from
c2681db to
439c4b3
Compare
kostandinang
force-pushed
the
ka/fe-843-toolchain-profiles
branch
from
c5efb9a to
8837e9b
Compare
This was referenced Jun 15, 2026
PR SummaryMedium Risk Overview
A new Reviewed by Cursor Bugbot for commit b44eaa1. Bugbot is set up for automated code reviews on this repo. Configure here. |
kostandinang
force-pushed
the
ka/fe-843-toolchain-profiles
branch
from
8837e9b to
b84fbda
Compare
kostandinang
force-pushed
the
ka/fe-864-orchestrator-enhancements
branch
from
95ac829 to
eb6e902
Compare
kostandinang
force-pushed
the
ka/fe-843-toolchain-profiles
branch
from
b84fbda to
e4e3afe
Compare
kostandinang
force-pushed
the
ka/fe-864-orchestrator-enhancements
branch
2 times, most recently
from
39d50cb to
bb36ee8
Compare
kostandinang
force-pushed
the
ka/fe-843-toolchain-profiles
branch
from
e4e3afe to
8557ba7
Compare
kostandinang
changed the title
kostandinang
changed the base branch from
ka/fe-843-toolchain-profiles
to
graphite-base/212
kostandinang
changed the base branch from
graphite-base/212
to
ka/fe-841-pi-sdk-embed
… port + assembles URLs (slice 2) Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a
… probe runProbe's readiness poll and feature fetch used bare global fetch with no timeout. A server that accepts a connection but never responds would block await fetch forever — the wall-clock READY_TIMEOUT_MS is only checked between poll attempts, so it never fired, hanging the probe and the whole cook. Each fetch now carries a per-call AbortSignal.timeout; timeouts are overridable so the no-hang behavior is unit-tested fast. Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a Co-authored-by: Amp <amp@ampcode.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…he verify-epic verdict Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a Co-authored-by: Amp <amp@ampcode.com>
…able ProbeGrounder Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a Co-authored-by: Amp <amp@ampcode.com>
…d> via git plumbing Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a
Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a Co-authored-by: Amp <amp@ampcode.com>
runCook reads opts.dir raw — the launch-cwd default lives only in parseCookArgs, which serve bypasses. With dir:'' cook resolved the just-emitted plan path against process.cwd() and would clone '' for brownfield, so serve only worked when greenfield and process.cwd()===launchCwd===project root. serveCookOptions/runServe now take the resolved cook dir; cli passes launchCwd (the same dir plan writes the plan to). Corrects the stale test that asserted cook.dir===''. Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a Co-authored-by: Amp <amp@ampcode.com>
- I126-K: name the shared runVerification seam (FE-872); evaluateVerificationTargets is deleted - decision 166: brownfield promotion is no longer a follow-on — landed as decision 168 (FE-877) - decision 168 (new): brownfield auto-promotion (plumbing-only) + brunch serve capstone (FE-877/878) - I128-K: --out is greenfield-only (brownfield auto-promotes) - I135-K (new): brownfield promotion never touches the user's checkout Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a Co-authored-by: Amp <amp@ampcode.com>
The serve and plan branches duplicated the spec gate verbatim (resolveBrunchProject -> createDb -> existence check -> snapshot -> completeness assert -> db close + uniform error). withCompletedSpec now owns it; parsing is a thunk so parse errors report through the same 'Failed to run brunch <command>' channel. Pure refactor — cli.test.ts green. Amp-Thread-ID: https://ampcode.com/threads/T-019ecb9a-9a08-733b-833d-76885fc8243a Co-authored-by: Amp <amp@ampcode.com>
Introduces a single emit(CookEvent) presentation boundary so terminal
output stops being smeared across console.error/log() in the orchestrator
CLI. Foundation: presenter.ts root + presenter/{events,bus,select,plain,
silent}.ts.
- selectPresenter(command,isTTY,ci,reporterFlag): pure decision table →
plain (CI/non-TTY/default) | silent (agent, keeps stdout JSONL-clean) |
ink (interactive TTY; falls back to plain until slice 2).
- CookBus: synchronous fan-out; a thrown presenter is downgraded to a
process warning so presentation can never abort a run.
- PlainPresenter: CookEvent → stderr, byte-exact for the plan arms; sink
injectable for the golden differential.
- plan-runner migrated to emit CookEvents; cli.ts plan/serve wired through
createCookBus. cook left untouched (still behavior-preserving).
Oracle per SPEC I136-K: plan-runner.test.ts now drives a capturing bus and
asserts the same stderr; npm run verify green.
Slice 1b (cook surface + injected-clock elapsed timer) queued in
memory/CARDS.md.
Co-Authored-By: Claude <noreply@anthropic.com>
Routes cook/serve terminal output through the emit(CookEvent) boundary,
completing the seam across all three commands.
- cook-cli: banner / completion summary / promotion / petrinaut blocks and
the early-exit diagnostics now emit {kind:'line'} through the bus; the
petrinaut-setup log is bus-backed. runCook takes a bus (defaults to
createCookBus('cook')); serve shares one bus across plan+cook.
- pi-actions: per-action log()/logVerbose() become structured action/
verbose CookEvents; the module is now console-free. The module-level
Date.now() elapsed timer is gone — the presenter owns it.
- PlainPresenter: gains an injected clock (I136-K). A cook-start event
seeds runStart; the elapsed prefix is computed at render time, so the
cook surface now has a deterministic byte-exact golden.
Verified: presenter goldens (plan + cook arms incl. fake clock),
brownfield-smoke runs cook end-to-end through the bus, npm run verify green.
ink still falls back to plain — that's slice 2.
Co-Authored-By: Claude <noreply@anthropic.com>
Makes the `ink` backend real (it no longer falls back to plain on a TTY).
- format.ts + clock.ts: line formatting + the elapsed clock extracted from
PlainPresenter so the plain and Ink backends share one formatter and can't
drift. PlainPresenter is now a thin sink over formatCookEvent.
- phase.ts: nextPhase — a pure, monotonic brigade tracker
(prep→recipe→cook→taste→plate→serve) projected from the event stream.
Coarse for now (post-hoc events); precise in-flight transitions are 2b.
- run-store.ts: folds CookEvents into { phase, lines } with a stable
snapshot for useSyncExternalStore.
- ink/: egg-logo.ts (ANSI mark), app.tsx (egg header + brigade strip +
bounded activity log), ink-presenter.tsx (renders to STDERR; stdout stays
reserved). makePresenter('ink') now returns InkPresenter.
Adds ink@^7 + ink-testing-library@^4 (React 19.2 satisfies the peer dep).
Verified: phase/run-store units, ink-testing-library frame (egg + active
phase + activity line), non-TTY path still plain (brownfield-smoke), full
build bundles the tsx. Real-terminal walkthrough is outer-loop debt; the
dead-air waiting fix is slice 2b.
Co-Authored-By: Claude <noreply@anthropic.com>
Closes the dead-air problem: long waits now show what brunch is doing. - events: activity-start / activity-progress / activity-end. - pi-actions: runPi self-brackets every agent session (start → finally end) with a throttled KB heartbeat off its token stream; the test-run and probe waits bracket via a small withActivity helper. All close in finally, so a spinner can't hang — covered by a test that fails the session mid-wait. - cook-cli: promotion brackets via a `promoting` helper. - run-store: a pending map (start adds, progress updates detail, end removes); activity events stay out of the scrolling log. - ink: PendingPanel renders a live spinner + label + elapsed + detail, with a tick interval that runs only while something is pending. Plain/CI prints one `⋯` start line per wait. Known limit: test-runner uses blocking spawnSync, so the spinner freezes (but stays labeled) during a test run; the async pi session animates. Real-terminal walkthrough is outer-loop debt. Verified: run-store pending units, ink frame (panel shows/clears), balanced brackets incl. on session failure, npm run verify green. Co-Authored-By: Claude <noreply@anthropic.com>
ln-review caught that nothing ever called bus.dispose() — harmless for plain/silent, but on a real TTY the Ink app was never unmounted, so `brunch cook`/`serve` would hang after the run. - withCookBus(command, fn): builds the bus, runs the work, and disposes it (→ unmounts Ink) in finally. One owner, no split ownership. - runCook takes a required bus (drops the in-cook createCookBus default). - cli.ts cook/plan/serve paths run through withCookBus; serve's single shared bus is disposed once after the cook stage. Verified: withCookBus disposes on success and on throw (spy); CookBus.dispose fan-out test stands; npm run verify green. Remaining real-terminal debt is now purely visual. Co-Authored-By: Claude <noreply@anthropic.com>
) The cook banner/summary text had no oracle (the migration preserved it verbatim but nothing guarded against drift). Extract cookBannerLines / cookSummaryLines as pure functions and golden-test them; runCook feeds their output to the bus. Covers completed + halted runs incl. the epic/slice tree. npm run verify green. Co-Authored-By: Claude <noreply@anthropic.com>
…arks Per feedback: drop the egg, use the "brunch" wordmark tinted with the brunch.ai brand gradient (HASH blue→indigo→violet, one hex per letter), and keep the brigade/status glyphs as the original monochrome marks (✓ ◐ ○) rather than emoji. egg-logo.ts → wordmark.ts. Plain/CI backend stays untinted. Ink frame tests updated. Co-Authored-By: Claude <noreply@anthropic.com>
The panel re-rendered every 120ms recomputing toFixed(1) elapsed, so the number jittered at the decimal. Add formatElapsed (whole seconds under a minute, m:ss above), use it in the panel, and slow the spinner tick to 250ms. The static action-log prefix (a fixed record) keeps its one-decimal form. Co-Authored-By: Claude <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
kostandinang
force-pushed
the
ka/fe-864-orchestrator-enhancements
branch
from
38400fc to
b44eaa1
Compare
kostandinang
force-pushed
the
graphite-base/212
branch
from
b75cac0 to
977ea9c
Compare
lunelson
approved these changes
Jun 19, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stack Context
This is the middle layer of the cook/brunch-serve stack — the FE-864 "orchestrator improvements" umbrella. It builds the end-to-end brownfield delivery pipeline on top of the pi-agent foundations (#194): take a completed spec, detect the repo's toolchain, run the cook agent, verify the result against a live app, promote it, and drive the whole thing from one
brunch servecommand. The operational-hardening PR (#224) stabilizes what this introduces.Consolidates the FE-864 umbrella and its children (FE-867, FE-871, FE-872, FE-875, FE-876, FE-877, FE-878, FE-879) into one PR.
What?
The brownfield delivery pipeline, in dependency order:
plan/cook/servewith brigade names as phases.detectProfile, monorepo-robust test-dir + workspace-runner detection, generated tests co-located in the repo's own test dir, fail-loud on ambiguous evidence; wired into plan emission.buildProbeSpec(port allocation + URL assembly) with bounded, strict-deadline probe calls.ProbeGrounder(Half B).cook/<runId>via git plumbing.node_modules; fail loud on slice-id/parent collisions.Why?
Foundations (#194) made the cook agent safe and portable to run; this layer makes it deliver. Brownfield delivery from a spec needs the orchestrator to understand an unfamiliar repo (detection), tell real failures from infra noise (classification), confirm the result actually runs (probe + integration oracle), land it without clobbering the working tree (promotion onto
cook/<runId>), and present all of it coherently from a single command (brunch serve). FE-879 keeps per-slice runs cheap so the pipeline scales.Folded from the FE-864 umbrella down through its linear children, with FE-879 (a side branch off the umbrella) folded in.