Conversation
Replace requirements*.txt with a standard pyproject.toml, target Python 3.13 (matching Pyodide 0.29.x), and add mise.toml plus uv-based dev loop. Updates Dockerfile, pytest.ini, and .gitignore accordingly.
…le-plot emission Add ProcessOptimizer-backed multi-objective optimization with pareto plots and per-dimension JSON single plots, and extract plot emission into a dedicated plot_emitters module. Includes the optimizer refactors that split process_result into focused helpers.
Add the pickled_state module with request fingerprinting and pack/unpack-if-valid round-trip, wire the handler to consume a pickled model cache (surfacing pickledUsed) and a selectedPoint override with fall-through warnings, and document both in the OpenAPI extras schema.
Use secrets.compare_digest for API-key comparison, stop logging raw access tokens to stdout, and refuse the default AUTH_API_KEY='none' in production.
Replace print() with central logging configuration, and add a types module with TypedDicts for the request, response, and cache payloads behind a permissive mypy baseline.
Add an end-to-end pareto HTTP suite plus unit coverage for auth, pickled_state, plot_emitters, and types; introduce a conftest with a stable PICKLE_KEY; enforce no print() under optimizerapi/; expand the optimizer tests; and add sample curl scripts. Drops the dead tests/context.py module.
Add AGENTS.md (with CLAUDE.md pointing to it) as the canonical setup, workflow, and style reference; add an end-to-end API usage guide and current-context handoff notes; refresh the README; and capture the 5-phase audit and pareto-extras redesign plans/spec under docs/.
Establish ADRs under docs/adr/ as the only durable design doc kept on main, and add a vendor-neutral branch-finalize workflow. - docs/adr/: Nygard-style template, index/README, and two records — 0001 (pickled model as an advisory, fingerprint-validated cache hint, distilled from the pareto-extras redesign) and 0002 (the ADR / branch-only agent-docs convention itself, with the alternatives weighed). - AGENTS.md: vendor-neutral 'Decisions and finishing a branch' section — ADR policy plus the advisory branch-finalize flow (distil a decision into an ADR, keep plans/specs through review, remove them in a cleanup commit before merge). Surfaced by the agent, not mechanically enforced. - CLAUDE.md: points at the finish-branch skill and the new AGENTS section. - .claude/skills/finish-branch: encodes the flow, adapted to this repo's pytest/flake8/mypy checks; .gitignore narrowed to track skills while keeping local Claude settings ignored. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The workflow pinned Python 3.9 and installed deps from requirements-freeze.txt, which the pyproject migration removed — so runtime deps (rq, cryptography, numpy, ProcessOptimizer) never installed and optimizerapi/types.py failed to import (NotRequired needs 3.11+). Replace setup-python with jdx/mise-action so CI provisions the same Python 3.13 + uv as local dev (mise.toml) and installs the project via 'uv pip install -e .[dev]'. Add a mypy step (now matches the AGENTS.md claim) and a .flake8 config excluding the local .venv/env so 'flake8 .' lints only project sources in CI and locally. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The image still targeted python:3.9-bullseye while pyproject now requires >=3.13, and the builder copied only pyproject.toml — so 'pip install .' could neither run on 3.9 nor build the project (setuptools needs README.md and the optimizerapi/ source). Rebuild the Dockerfile on python:3.13-slim-bookworm, install with uv (matching dev/CI), and copy the metadata + source the build actually needs, with a uv cache mount so the ProcessOptimizer git dependency isn't re-fetched each build. Workflow: build linux/amd64 only (arm64 was emulated under QEMU and dominated build time), add GHA layer caching (cache-from/to: type=gha), and bump the docker/* and checkout actions to current major versions. Add a .dockerignore so the local .venv (multi-GB) and other cruft stay out of the build context. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
torch arrives transitively via ProcessOptimizer and, from PyPI, drags in the full NVIDIA CUDA stack (~5 GB: cublas, cudnn, nccl, cusolver, ...). The API only does CPU inference, so this was pure build-time and image-size cost. Pin torch to PyTorch's CPU index via [tool.uv.index] + [tool.uv.sources]. uv source pins apply only to *direct* dependencies, so torch is also declared explicitly in [project.dependencies]. Resolution now yields torch==2.12.0+cpu with zero CUDA packages (88 -> 69 packages). Verified by building the image and booting it: GET /v1.0/health -> 200, /v1.0/ui/ -> 200, image 1.42 GB (was ~6 GB+), torch-2.12.0+cpu installed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
current-context.md was a temporary design-checkpoint handoff (2026-05-18) that self-documented 'Delete after the brainstorm produces an updated plan'. That brainstorm happened (docs/superpowers/.../pareto-extras-redesign*), the work landed, and the durable decision is now captured in ADR 0001. The doc also referenced things that no longer exist (.sisyphus/, result.json) and stale facts (29 tests, untracked CLAUDE.md). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Per the ADR convention (ADR 0002 / AGENTS.md), agent-generated plans and specs
are branch-only working artifacts kept through review and removed before merge.
The durable decision from the pareto-extras redesign is captured in ADR 0001;
the 5-phase audit plans' outcomes live in the code and commit history. Content
remains recoverable from git history.
Removed:
- docs/superpowers/plans/2026-05-18-audit-{overview,phase-1..5}.md
- docs/superpowers/plans/2026-05-18-pareto-extras-redesign.md
- docs/superpowers/specs/2026-05-18-pareto-extras-redesign-design.md
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
1 participant
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.
No description provided.