From dfeec8a88bc35c6583bbeb12ff3ffdc1b6b70ebc Mon Sep 17 00:00:00 2001 From: garrytan Date: Sun, 1 Nov 2026 14:44:41 +0000 Subject: [PATCH] docs: outside-voice default and voice-models.json pin; Step 14.5 runs once on Sonnet (GSTA-167 PR B) Adds the prb-docs-2 auditor's docs/unattended.md paragraph and fixes capy/skill/gstack/SKILL.md:49 at the same byte count (22,365 B), so the 5,312-token context-budget ceiling is unchanged. The ship-driver test pins the new 14.5 wording instead of the dropped haiku aside. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> --- capy/skill/gstack/SKILL.md | 2 +- docs/unattended.md | 8 ++++++++ test/ship-driver.test.ts | 2 +- 3 files changed, 10 insertions(+), 2 deletions(-) diff --git a/capy/skill/gstack/SKILL.md b/capy/skill/gstack/SKILL.md index 12e9d01a2..5685cc63b 100644 --- a/capy/skill/gstack/SKILL.md +++ b/capy/skill/gstack/SKILL.md @@ -46,7 +46,7 @@ Capy checklist for `/autoplan`: drive `gstack-autoplan`, never a hand adaptation Token caps on /ship: pass a brief's cap as `--token-cap `. /ship reads `cost.tokens` and `cost.credits` from `tools.thread.get({ thread_id: "" })` (its own jam, subagents included; `"this"` names the root thread, so only a root run may pass it) inside the same execute program as the `gstack-ship-budget` command that takes them as `--reading --credits `, never in a call of its own. `ship/DRIVER.md` starts `--mode observe`: each program ends with `note`, and only a note past the cap stops (`docs/ship-budget.md`). -/ship on Capy is about a dozen calls, each one execute program (`ship/DRIVER.md`). Each review pass and Step 11 is one call to the checked-in pass program (fill only `input`): `const t0 = await now(); const p = (await tools.capy.bash({ command: ". '' && gstack-review-dispatch program", description: "Pass" })).output.split("\n"); return p[0] + "\n" + (await new (Object.getPrototypeOf(async () => {}).constructor)("tools", "sleep", "now", "input", JSON.parse(p[1]))(tools, sleep, now, { startedAt: t0, launch: "", cwd: "", base: "", out: "/p", pass: "-p", laneModel: "", planFlags: [], logStart: ["review"], jam: "", record: "", step: "9.2" }));` It plans, logs `logStart` (`reservation` mode from `RESERVATION_ID:`), sends each brief on `machine: "shared"`, then polls: one `tools.subagent.list` per ~20 s tick, returning 200 s after the program's first line, where a reader is settled only at status `done` or `failed` or with a `summary` (a just-created `idle` reader has not run); it starts readers under `rejected` with `tools.subagent.start` as slots free and stops on one stuck 20 minutes. It ends on `NEXT:` (`poll`: rerun) or `STOP:`. `models` maps roles to models (`outside-voice`: another family); Step 11 adds `stage: "voices"`; a full repeat pass, `diffBase: ""`. Steps 7, 8, 9.2.1 and 14.5 (`step: "14.5"`) use `extras`, JSON `{name, prompt_file, first, model}` (prompts verbatim from their sections, Step 7 `first`); they cannot block, so `model: { model_id: "anthropic/claude-sonnet-5-5", reasoning_mode: "low" }` (`anthropic/claude-haiku-5-5` also fits). Waits: never let 5 minutes pass between calls. While readers run, never end the turn on `wait`; rerun the loader. Otherwise re-issue `wait` with `duration_seconds: 200` and operation ids. At Step 22 report the `CALLS:` line of `gstack-ship-budget calls --events ` on your jam's transcript since /ship began. A small waiting child saves nothing. +/ship on Capy is about a dozen calls, each an execute program (`ship/DRIVER.md`). Each review pass and Step 11 is one call to the checked-in pass program (fill only `input`): `const t0 = await now(); const p = (await tools.capy.bash({ command: ". '' && gstack-review-dispatch program", description: "Pass" })).output.split("\n"); return p[0] + "\n" + (await new (Object.getPrototypeOf(async () => {}).constructor)("tools", "sleep", "now", "input", JSON.parse(p[1]))(tools, sleep, now, { startedAt: t0, launch: "", cwd: "", base: "", out: "/p", pass: "-p", laneModel: "", planFlags: [], logStart: ["review"], jam: "", record: "", step: "9.2" }));` It plans, logs `logStart` (`reservation` mode from `RESERVATION_ID:`), sends each brief on `machine: "shared"`, then polls: one `tools.subagent.list` per ~20 s tick, returning 200 s after the program's first line, where a reader is settled only at status `done` or `failed` or with a `summary` (a just-created `idle` reader has not run); it starts readers under `rejected` with `tools.subagent.start` as slots free and stops on one stuck 20 minutes. It ends on `NEXT:` (`poll`: rerun) or `STOP:`. `models` maps roles to models (`outside-voice`: another family, pinned in `voice-models.json`); Step 11 adds `stage: "voices"`; a full repeat pass, `diffBase: ""`. Steps 7, 8, 9.2.1 and 14.5 (`step: "14.5"`) use `extras`, JSON `{name, prompt_file, first, model}` (prompts verbatim from their sections, Step 7 `first`); they cannot block, so `model: { model_id: "anthropic/claude-sonnet-5-5", reasoning_mode: "low" }` (14.5 runs once). Waits: never let 5 minutes pass between calls. While readers run, never end the turn on `wait`; rerun the loader. Otherwise re-issue `wait` with `duration_seconds: 200` and operation ids. At Step 22 report the `CALLS:` line of `gstack-ship-budget calls --events ` on your transcript since /ship began. A small waiting child saves nothing. ## Plan, approve, ship diff --git a/docs/unattended.md b/docs/unattended.md index be924d334..98ab440c7 100644 --- a/docs/unattended.md +++ b/docs/unattended.md @@ -228,6 +228,14 @@ credits outside coverage. A parent reads the rows with `gstack-review-read` and treats `outside_status: unavailable` as missing coverage, not as a clean outside review. +The pass program (`lib/capy-review-pass.js`) picks the outside voice's model +when `input.models['outside-voice']` is empty and the lane model is +`anthropic/`: `openai/gpt-6.1-sol` at medium reasoning, the pairing the no-loss +rounds ran. Any other lane must name a model. An outside voice from the same +provider as the lane model or the native voice stops the pass (`STOP:`) before +either voice is created, including on a rerun, and each outside voice prints an +`OUTSIDE_VOICE:` line saying whether it is the certified configuration. + ### The carry across a base merge or a test-only fix When the parent merges the base in or fixes a test after the receipt was diff --git a/test/ship-driver.test.ts b/test/ship-driver.test.ts index 33a0f3c9e..b5ad4472c 100644 --- a/test/ship-driver.test.ts +++ b/test/ship-driver.test.ts @@ -226,7 +226,7 @@ describe('variant G: the call plan, observe budget, waits and audit readers (GST expect(flat(doc)).toContain('`Coverage:` and Step 8\'s line come from `--audit` when no reader ran'); const capy = read('capy/skill/gstack/SKILL.md'); expect(capy).toContain('model: { model_id: "anthropic/claude-sonnet-5-5", reasoning_mode: "low" }'); - expect(capy).toContain('`anthropic/claude-haiku-5-5` also fits'); + expect(capy).toContain('(14.5 runs once)'); }); // Value: protects=a driven /ship dispatches every blocking reader on its rendered brief file unchanged, checks it with `sent`, merges findings only through `collect`, gates judgment with `judge` and verifies the run before the bind; fails_when=the driver lets the model write or trim a brief again or merge findings in prose (round 4b: arm D lost 8 blockers with 75 of 112 briefs paraphrased);