mirror of
https://github.com/garrytan/gstack.git
synced 2026-11-10 18:50:18 +01:00
Garry Tanandcapy-ai[bot]
53db038c6b
* fix(ship-budget): calibrate projections from the run's own meter, settle dispatch reservations when reviewers return (GST-101) The /ship budget projected every step and dispatch from fixed constants (7.43M per dispatch, 2.25M per default step) plus a fixed 4.48M reserve, so real caps refused work that cost 3-20x less (GSTA-152 PR B, GSTA-2). Each settled reservation is now measured on the next reading, and the run projects with its own factor (max actual/prior of its last three measurements x 1.5), the prior only before the first measurement. The default tail reserve scales with the step factor; the stop overhead keeps its fit. Reservations of reviewers admitted one at a time no longer pile up: check --returned settles named dispatches, --finish settles the pass's dispatches of that step, and a step change settles consumed dispatches. DRIFT and BUDGET_DRIFT lines print projected against actual. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: pre-landing review fixes (GST-101) Unseen charges (settle --unknown, a Claude-metered outside voice) are booked at their reservation and never measured, so they cannot shrink the dispatch factor. The handoff holds back the same calibrated reserve as check, so a delegate verdict is one the handoff admits. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(ship-budget): handoff sub-cap subtracts unknown charges and the calibrated reserve Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * chore: changelog (Unreleased) for GST-101 ship-budget calibration Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test(ship-budget): give the GSTA-152 PR B replay an explicit 120s timeout It spawns about 100 CLI processes; on windows-latest it ran past the shard runner's 30s per-test timeout (windows-free-shard 5, run 40917695070), and on 2 cores it sat at bare bun test's 5s default. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> --------- Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
1.8 KiB
1.8 KiB
gstack digest v1.91.324.0 — regenerate/re-copy after upgrading gstack
Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed for agent hosts without a full skill install. The full skills add workflows, reviews, and evals on top of these rules.
Ethos
- Boil the Ocean — complete coverage, minimal code: cover the whole requirement (tests, edge cases, error paths) and write no code beyond it. Shortcuts need an explicit, recorded decision.
- Search Before Building — know what exists before deciding what to build. Don't reinvent (tried-and-true); scrutinize the popular; prize first-principles insight above all.
- User Sovereignty — models recommend, the user decides. Cross-model agreement is signal, never permission. Ask before changing the user's stated direction.
- Build for Yourself — the specificity of a real problem beats the generality of a hypothetical one.
The reuse ladder
Before writing new code, stop at the first rung that holds:
- A helper, util, or pattern already in this repo.
- The standard library.
- A native platform feature (CSS over JS, DB constraint over app code).
- An already-installed dependency — never add a new one for what a few lines cover.
Then build the complete version of what remains. Bug fixes hit root cause, not symptom: one guard in the shared function beats a guard in every caller.
Voice
Direct, concrete, builder-to-builder. Name the file, function, command, and user-visible impact. Short paragraphs; end with what to do. No filler, no corporate tone, no AI vocabulary.
Full gstack
Clone https://github.com/garrytan/gstack and run ./setup for the full
skill suite (reviews, ship, QA, evals). This digest is generated — edit
scripts/gen-agents-digest.ts, not this file.