Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
97 changes: 57 additions & 40 deletions HANDOFF.md
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
# Handoff

_Last updated: 2026-08-14 02:06 KST by Codex_
_Last updated: 2026-08-14 02:34 KST by Codex_

## Goal

Expand All @@ -13,25 +13,24 @@ _Last updated: 2026-08-14 02:06 KST by Codex_

## Current Status

- `weightclass 0.13.0` is merged, tagged, published, and installable.
- Release commit and annotated `v0.13.0` tag target:
`155f7446b295f3d226c0a392c3b52d93a4b5644f`.
- GitHub Release: <https://github.com/ictechgy/weightclass/releases/tag/v0.13.0>.
- Final release-gating CI:
[31686174951](https://github.com/ictechgy/weightclass/actions/runs/31686174951), completed successfully.
- `weightclass 0.14.0` is merged, tagged, published, and installable.
- Release commit and annotated `v0.14.0` tag target:
`7860d0ba3aed3f43f4de3401fb137e0114a5cfcf`.
- GitHub Release: <https://github.com/ictechgy/weightclass/releases/tag/v0.14.0>.
- Release-candidate PR CI runs
[31724967486](https://github.com/ictechgy/weightclass/actions/runs/31724967486)
and [31724971200](https://github.com/ictechgy/weightclass/actions/runs/31724971200)
completed successfully.
- Immutable Release workflow:
[31686408438](https://github.com/ictechgy/weightclass/actions/runs/31686408438), completed successfully.
- PyPI: <https://pypi.org/project/weightclass/0.13.0/>.
- `0.13.0` adds tier-specific opaque Grok model overrides to packaged
`review-preset`, `route`, and `run`, together with the repository
security/performance and release hardening described below. Custom selections
remain explicitly unqualified; built-in routes and provider boundaries are
unchanged.
- The merged `0.14.0` release candidate adds the cost-recommendation feature:
[31725291232](https://github.com/ictechgy/weightclass/actions/runs/31725291232), completed successfully.
- PyPI: <https://pypi.org/project/weightclass/0.14.0/>.
- `0.14.0` adds the cost-recommendation feature:
task-free `review-cost-profile` plus non-executing `recommend`, together with
aggregate-only sanitized provider-usage evaluation. It is not tagged,
published, or installed through Homebrew yet. Built-in routes and every
aggregate-only sanitized provider-usage evaluation. Built-in routes and every
execution path remain unchanged.
- `0.13.0` is an earlier published release. It added tier-specific opaque Grok
model overrides to packaged `review-preset`, `route`, and `run`; custom
selections remain explicitly unqualified.
- An approved exact-command measurement probe used one already-public standard
fixture and exactly six vendor calls: built-in standard/medium versus the
packaged standard/low candidate for Codex, `agy`, and Grok. All six commands
Expand Down Expand Up @@ -103,9 +102,9 @@ _Last updated: 2026-08-14 02:06 KST by Codex_
output. A passing fixed-subscription quota result is `capacity_only`, never a
monthly-bill claim or cost-recommendation input. A nine-pair canary remains a
scored `no-go`; the fixed 30-pair and slice/quality gates are unchanged.
- The source-of-truth Homebrew formula was updated on `main` by
`e18fd3988a6fa6ad642f44a084b436c6507a2f6b`. The matching tap commit is
[`1244ed6`](https://github.com/ictechgy/homebrew-tap/commit/1244ed6f0baf6cf2c557bec356230e3ea199961d).
- The source-of-truth Homebrew formula and public tap install the canonical
PyPI 0.14.0 sdist. The matching tap commit is
[`7d393d0`](https://github.com/ictechgy/homebrew-tap/commit/7d393d0a3d41405c883e48a6bb53197473c6c600).
- `v0.8.1` was never published to PyPI: its Release workflow failed only on
Ruff import ordering. It had no GitHub Release and was deliberately deleted
from both the local and `origin` tag namespaces. Do not recreate it.
Expand Down Expand Up @@ -373,7 +372,7 @@ _Last updated: 2026-08-14 02:06 KST by Codex_
reviewed action metadata runs on Node 24, replacing the Node 20-based v5
generation that GitHub runners warned about.

## Current Unreleased Cost Recommendation
## Completed in 0.14.0

- Added `src/weightclass/cost_recommendation.py` with strict version-1 cost
profile and qualification-card schemas. Inputs are bounded regular JSON,
Expand Down Expand Up @@ -487,6 +486,30 @@ _Last updated: 2026-08-14 02:06 KST by Codex_

## Verification

- Fresh `0.14.0` release verification on 2026-08-14:
- PR #35 merged the provider-usage gate and rejected-candidate safeguards;
PR #36 merged the single version bump. Both PR push/merge CI runs passed
Python 3.10-3.14, macOS 3.10/3.14 boundaries, lint, strict typing, build,
metadata, and extracted-sdist isolation.
- Local Python 3.10.20 and Python 3.14.6 full suites passed 816 tests each
with `ResourceWarning` as an error. Ruff 0.16.2, strict mypy over 107 source
files, compileall, strict Twine metadata, and `git diff --check` passed.
- The Python 3.13 build produced exactly the 0.14.0 wheel and sdist. Extracted
sdist isolation passed 810 tests with 13 platform skips, and a clean wheel
install passed version, classification, and recommendation parser smokes.
- Release workflow `31725291232` built one immutable candidate, revalidated
it on Python 3.10/3.14 and macOS 3.10/3.14, and published it through PyPI
Trusted Publishing after the required environment review.
- PyPI exposes exactly one wheel and one sdist, neither yanked. Independent
downloads matched SHA-256: wheel
`68262b12b7007a580201e0e37e2fd3edb0e2555f077b096519bf1deff8180cbc`;
sdist `051345050e467455d8c2eedf963b690a4e49dfe1c1219ea6541e2f8be3b68d97`.
The downloaded wheel reported `weightclass 0.14.0` and passed the local
low-tier and `recommend` smokes.
- Formula-scoped style and strict audit, source reinstall, `brew test`, and
direct low/high classification plus recommendation parser smokes passed.
`/opt/homebrew/bin/wclass --version` reports `weightclass 0.14.0`.

- Fresh unreleased cost-recommendation verification on 2026-08-13:
- Behavior-first RED→GREEN covered missing `recommend`, flat-cost
abstention, configuration-before-task ordering, full qualification-card
Expand Down Expand Up @@ -747,11 +770,8 @@ _Last updated: 2026-08-14 02:06 KST by Codex_

## Blockers & Open Questions

- No known blocker or mandatory release/deployment step remains for `0.13.0`.
- The cost-recommendation and provider-usage work is merged to `main` by PRs
#34 and #35. The `0.14.0` release candidate still needs its immutable tag,
Release workflow, PyPI verification, and Homebrew formula update. Do not
republish immutable `0.13.0`.
- No known blocker or mandatory release/deployment step remains for `0.14.0`.
Do not republish or move its immutable version tag.
- GitHub repository settings were updated and re-read through the API on
2026-08-13. The active `Protect version tags` ruleset targets `refs/tags/v*`,
has no bypass actor, and blocks updates and deletions. The `pypi` environment
Expand Down Expand Up @@ -782,42 +802,39 @@ _Last updated: 2026-08-14 02:06 KST by Codex_

## Next Steps

1. Merge the `0.14.0` release-version change, run the full immutable release
gates, create the protected annotated `v0.14.0` tag, approve and verify the
PyPI publication, then update and test the Homebrew formula.
2. Keep the cost-focused Claude policy explicit opt-in only. Collect real-user
1. Keep the cost-focused Claude policy explicit opt-in only. Collect real-user
compatibility feedback without task telemetry, provider-price inference, or
a claim about actual bills. Do not change built-ins, standard/high routes,
schema 2, or posture vocabulary from the low-target result.
3. Keep the rejected Codex Terra-to-Luna low-model candidate out of automatic
2. Keep the rejected Codex Terra-to-Luna low-model candidate out of automatic
routing and cost recommendations. If testing a materially different future
candidate, bind only `--low-model` plus reviewed low effort, keep standard
and high unchanged, and require a fresh canary plus an exact packaged-route
qualification. Stop on any new critical failure; do not reuse the Luna
aggregate as evidence for another model or for subscription quota savings.
4. Treat a passing quota result as capacity evidence only. Only passing
3. Treat a passing quota result as capacity evidence only. Only passing
metered-cost evidence may feed a cost profile/qualification card, and even
then keep the exact qualified tier explicit opt-in. Evaluate Codex, `agy`,
and Grok independently; do not combine results or infer pricing.
5. Re-enable Linux Claude semantic triage only after reviewing a concrete
4. Re-enable Linux Claude semantic triage only after reviewing a concrete
filesystem-containment command and its process-tree boundary.
6. Preserve Protocol 1 compatibility, explicit cross-vendor opt-in, task
5. Preserve Protocol 1 compatibility, explicit cross-vendor opt-in, task
no-retention, and the single-reviewed-child boundary.

## Resume Prompt

Open `/Users/jinhongan/Desktop/subscription-agent-router`, read `HANDOFF.md`
and `AGENTS.md`, then continue from: `weightclass 0.13.0 remains the immutable
published release. The 0.14.0 release candidate contains the merged same-vendor
advisory cost router: review-cost-profile plus recommend, strict cost/qualification
and `AGENTS.md`, then continue from: `weightclass 0.14.0 is the immutable
published and Homebrew-installed release. It contains the same-vendor advisory
cost router: review-cost-profile plus recommend, strict cost/qualification
schemas, full evidence fingerprints, fixed conservative gates, explicit
abstention, provider capability/task-delivery receipts, docs, and ten focused
tests. Built-ins and execution paths are unchanged. Python 3.10/3.14 each
passed 810 tests; Ruff, strict mypy, compileall, build, strict
Twine, extracted-sdist isolation, and a clean-wheel recommendation smoke are
green. Run the immutable 0.14.0 release process before publishing; do not
republish 0.13.0. The merged provider-cost follow-up keeps unqualified standard
routes at medium, rejects no-op cost candidates, and adds sanitized provider-export
green. Do not republish or move v0.14.0. The merged provider-cost follow-up keeps
unqualified standard routes at medium, rejects no-op cost candidates, and adds
sanitized provider-export
metered-cost/quota scoring. Its local full suites currently pass 816 tests on
Python 3.10 and 3.14. Do not start another provider run until the user supplies
the exact low-tier model label and approves a bounded external measurement
Expand Down
4 changes: 2 additions & 2 deletions packaging/homebrew/weightclass.rb
Original file line number Diff line number Diff line change
Expand Up @@ -7,8 +7,8 @@ class Weightclass < Formula

desc "Local, policy-driven routing for agent CLI workflows"
homepage "https://github.com/ictechgy/weightclass"
url "https://files.pythonhosted.org/packages/36/0f/5ea0f285941b7bd2711289b052ae973cd4d1df27878e4832c5c590793e71/weightclass-0.13.0.tar.gz"
sha256 "9ba9db753c4aaf5189bf0210cd7933bd29a68168fcf473aeb1ad302586297ae9"
url "https://files.pythonhosted.org/packages/5a/f4/cd3fce8b85e624881d4a7c071c9953fd66f6d8a2adef30d9357f4b9dcb40/weightclass-0.14.0.tar.gz"
sha256 "051345050e467455d8c2eedf963b690a4e49dfe1c1219ea6541e2f8be3b68d97"
license "MIT"

depends_on "python@3.13"
Expand Down