active.md

Active Tasks

benchmark-calibration

• issue: PyAutoLabs/autolens_assistant#59
• status: workspace-dev — first calibration campaign: teacher × {sonnet, haiku} + easy × {sonnet} via claude-code-subagent harness, serial (memory); records → benchmarks/runs/, RESULTS.md regen, rubric verdict on issue; PR at end
• autonomy: supervised effective (human-directed launch 2026-07-10 in-conversation, continuing #57 --auto chain; heart-ack: same set as #58 ship, in-session)
• note: judge = claude-fable-5 (this session) for judged rows; operator replies honest/minimal via SendMessage; failures recorded
• worktree: none (in-place: autolens_assistant on feature/benchmark-calibration)
• repos:
  • autolens_assistant: feature/benchmark-calibration

markdown-example-renderings

• issue: PyAutoLabs/PyAutoBuild#134
• status: library-dev — build generate_markdown.py (execute curated scripts → nbconvert md + PNGs) in PyAutoBuild, then autolens pilot (root start_here + imaging five + guides trio); phase 2 (other dataset types/workspaces/HowTo) files as a new prompt at ship
• autonomy: safe effective (--auto launched 2026-07-10; docs ≤ medium after phasing; plan on issue; NO heart-ack given — any Heart YELLOW at ship parks)
• note: execution never TEST_MODE; modeling images via completed-run resume cache (first build pays one real sampling run); images committed — verify .gitignore doesn't swallow markdown/ outputs
• worktree: ~/Code/PyAutoLabs-wt/markdown-example-renderings
• repos:
  • PyAutoBuild: feature/markdown-example-renderings
  • autolens_workspace: feature/markdown-example-renderings

preopt-breakdown-dashboard

• issue: PyAutoLabs/autolens_profiling#59
• status: workspace-dev — imaging leg DONE 5/5 clean; HEADLINE: real >=~1.8x library slowdown on mesh cells since May (identical cell configs; confirmed vs phase-3 quiet-machine runtime; F-matrix ~50% everywhere); alma_high cells on retry (root cause = missing dataset + auto-simulate OOM; seeded 234MB from phase-3 wt); laptop numbers = ambient-load fallback tier; A100 leg dispatch-ready (15 submits, 3bb613a) but RAL GPU nodes DOWN
• campaign: driver scratchpad/breakdown_campaign.sh (session a66a757a), logs scratchpad/breakdown_logs/; interruption-safe — each cell's JSON persists to results/breakdown/ on completion; cold-resume = rerun remaining cells' commands from the driver (skip cells whose v-current JSON exists)
• autonomy: supervised effective (--auto launched 2026-07-10; header supervised binds over maintenance cap safe; plan on issue; no heart-ack given — any Heart YELLOW at ship parks)
• claim-override: human-directed 2026-07-10 — proceeds alongside profiling-preopt-campaign's autolens_profiling claim ("i dont see a clash"); phase 4 works likelihood_breakdown/ + README, phase 3 works likelihood_runtime/ — distinct paths; coordinate merge order at ship (README/build tooling may touch both — rebase on campaign branch if needed)
• note: dashboard deliverable depends on phase-3 PreOptimizationTimes runtime baseline (not yet tagged); breakdown runs (deliverable 1) proceed independently; HPC CPU/A100 legs gated on RAL availability
• worktree: ~/Code/PyAutoLabs-wt/preopt-breakdown-dashboard
• repos:
  • autolens_profiling: feature/preopt-breakdown-dashboard

pj011646-wfc3-parity

• issue: PyAutoLabs/PyAutoReduce#25
• session: claude --resume cc79d958-a1aa-45cb-b088-bd6cae94aa86
• status: awaiting-input — parked at ship sign-off (gate: tests 202 pass / smoke n-a / review CLEAN / Heart YELLOW 6-reason set UNACKED); branch feature/pj011646-wfc3-parity committed locally bd9806b, NOT pushed; phases 0-2 verdicts on #25; model-parity fits parked (2 OOMs at n_live=100, laptop contended)
• autonomy: safe effective (--auto launched 2026-07-10; plan human-approved in-session pre-launch; no heart-ack given)
• note: keck-ao/slacs1430 pattern — analysis on PyAutoReduce main, NO worktree claim (jwst-frame-feasibility holds it); ship gated on claim release or human direction; autolens_assistant driver-only (in-place branch belongs to assistant-ref-mechanics — scratch writes only); model-parity fits SERIAL on quiet machine (~5.5GB)
• question: PyAutoLabs/PyAutoReduce#25 (comment)
• worktree: none (analysis on PyAutoReduce main; branch feature/pj011646-wfc3-parity only at ship)
• repos:

rect-adapt

• issue: PyAutoLabs/PyAutoArray#372
• session: claude --resume 4bf88e6b-682d-4590-906f-77b68d059b26
• status: MERGED 2026-07-10 (human-directed via kernel-cdf session, coordinated order #375→#374; issue #372 closed with verification upthread) — entry ready to retire to complete.md by its owning session
• pr: PyAutoLabs/PyAutoArray#375
• autonomy: supervised effective (--auto chain 2026-07-09; bug cap binds over safe header; ship sign-off will park with question per contract)
• heart-ack: same 6-reason set as the assistant-ref-mechanics entry (in-session 2026-07-09); binds to exactly that set, any new reason parks
• note: consumers of edges_transformed = plot/inversion.py pcolormesh + mesh_geometry/rectangular_rotated.py — both inherit the fix; mcmc-corner-smoke (nightly blocker) queued in planned.md on autofit_workspace/PyAutoFit claims
• worktree: ~/Code/PyAutoLabs-wt/rect-adapt
• repos:
  • PyAutoArray: feature/rect-adapt

slacs1430-acs-parity

• issue: PyAutoLabs/PyAutoReduce#17
• session: claude --resume cc79d958-a1aa-45cb-b088-bd6cae94aa86
• status: MERGED 2026-07-10 (0face12, squash; human-authorized in-session) — branches deleted; #17 stays OPEN for the parked model-parity leg only; entry retires to complete.md once that leg's verdict lands
• pr: PyAutoLabs/PyAutoReduce#23
• resume(model-parity leg): ~/venv/PyAuto/bin/python prototypes/slacs1430_parity_fit.py autoreduce then ... legacy from PyAutoReduce root; compare fit_summary_*.json; verdict on #17
• note: acquire-dupe bug prompt RETIRED (fixed on main by #18's is_direct_product; my run disclosed on PR#23 as HAP-family); PJ011646 WFC3 follow-up prompt queued (methodology notes on #17)
• autonomy: safe effective (--auto launched 2026-07-09; plan human-approved in-session pre-launch; no heart-ack given)
• note: keck-ao pattern — analysis on PyAutoReduce main, NO worktree claim (frame-products holds it); outputs gitignored scripts/output/; script commit + PR gated until claim releases; autolens_assistant is driver-only (public template, no commits, scratch only)
• worktree: none (analysis on PyAutoReduce main; branch feature/slacs1430-acs-parity only at ship)
• repos:

keck-ao-acceptance-checks

• issue: PyAutoLabs/PyAutoReduce#13
• status: parked-resumable (laptop reset) — check 3 done: plate-scale finding (adapter 9.942 wrong; 9.952 pre-2015/9.971 post; fix gated); check 4 fit CHECKPOINTED ~1h in (94MB checkpoint.hdf5); resume command on the issue's interim comment
• resume: re-run prototypes/b1938_lens_fit.py (Nautilus auto-resumes from output/b1938_keck_acceptance identity path); then report θ_E vs 0.45" on #13; then propose keck_ao.md parity appendix + epoch-aware native_scale adapter fix (present-and-wait)
• autonomy: default present-and-wait (no --auto); analysis read-only in prototypes/, doc/adapter edits gated on user review
• worktree: none (analysis on PyAutoReduce main; branch only if an edit is approved)
• repos:

refactor-post-phase3

• issue: PyAutoLabs/PyAutoReduce#8
• session: claude --resume be7cb926-7874-4cc2-8c05-64c9644a64d9
• status: MERGED 2026-07-09 (371721f, squash) — human-directed merge in the keck-ao session (user authorized in-conversation); main checkout returned to main; PyAutoReduce claim released. Entry ready to retire to complete.md by its owning session (branch deletion via repo_cleanup)
• pr: PyAutoLabs/PyAutoReduce#10
• notes: witnesses byte-identical both paths; 2 disclosed fix riders (CPDIS fobj, ePSF window +20) caught by baseline capture
• autonomy: default present-and-wait (refactor cap safe; no --auto given)
• worktree: released (was in-place)
• repos:

ep-analytic-updates-scope

• issue: PyAutoLabs/PyAutoFit#1337
• session: claude --resume 3589268b-e5c9-4b32-b655-d07f732ea300
• status: complete-pending-pickup — phase 6 done; implementation plan for all 4 WPs on #1338 (plan-only, human-directed no-implement); backlog anchor feature/autofit/ep_analytic_updates.md
• plan: PyAutoLabs/PyAutoFit#1338
• autonomy: supervised (--auto, launched 2026-07-08)
• worktree: none (read-only)
• repos:

ep-deterministic-reconcile

• issue: PyAutoLabs/PyAutoFit#1336
• session: claude --resume 3589268b-e5c9-4b32-b655-d07f732ea300
• status: awaiting-input — phase 5 complete; recommendation A (keep both, document trade-off, resurrect #1153 test) on #1336 pending decision
• question: PyAutoLabs/PyAutoFit#1336 (comment)
• autonomy: supervised (--auto, launched 2026-07-08)
• worktree: none (read-only)
• repos:

ep-priors-fable-reassess

• issue: PyAutoLabs/PyAutoFit#1330
• session: claude --resume 3589268b-e5c9-4b32-b655-d07f732ea300
• status: awaiting-input — phase 0 complete; decision hub PyAutoFit#1331 open for maintainer/contributor guidance (fix-batch + 5 decisions)
• question: PyAutoLabs/PyAutoFit#1331
• worktree: none (read-only reassessment on PyAutoFit main @ 0f26ff2d8; verdicts land in PyAutoMind bug/priors)
• repos:

profiling-preopt-campaign

• issue: PyAutoLabs/autolens_profiling#56
• status: workspace-dev
• autonomy: supervised (--auto, launched 2026-07-08; local-CPU leg, RAL down)
• campaign: local-CPU matrix in flight (background); interruption-safe — resume with sweep.py --skip-gpu --skip-existing (dense, then --sparse imaging pass), then aggregate.py; full cold-resume steps on the issue (comment of 2026-07-08 evening)
• worktree: /home/jammy/Code/PyAutoLabs-wt/profiling-preopt-campaign
• repos:
  • autolens_profiling: feature/profiling-preopt-campaign

morning-status-release-rehearsal

• issue: PyAutoLabs/PyAutoHeart#39
• session: claude --resume ff9a8b2f-fda0-4bab-8962-1814843aa374
• status: RESOLVED 2026-07-09 (by session 5d58ef6a) — user re-set BOTH May-17 secrets (webhook + CLAUDE_CODE_OAUTH_TOKEN); morning_health dispatched → Slack POST success; digest needed 3 more CI fixes on Mind main (checkout, show_full_output, allowedTools Write; 51e869e/d042289/0b78d5f) → fully green, delivered. Details: issues/39#issuecomment-4924031684. Entry ready to retire to complete.md by its owning session
• prs: PyAutoBuild#119 + PyAutoHeart#40 + PyAutoMind#41 (independent; merge Mind last is tidiest — its morning_health reads the others)
• post-merge: dispatch morning_health.yml on Mind main (Slack POST leg); flip vars.RELEASE_MODE=live on PyAutoBuild when satisfied (human)
• autonomy: human-required effective (release cap; --auto launched 2026-07-08, plan approved in-session; ship sign-off + merge human)
• cleanup 2026-07-09: worktree removed + feature branches (local+remote) deleted via /repo_cleanup — all PRs were merged; remaining leg (webhook secret + morning_health.yml dispatch) is human-only and needs no repo claim

multi-shared-state-examples-phase-1-design

• issue: PyAutoLabs/PyAutoLens#599
• session: (this CLI session, 2026-07-10)
• status: library-dev
• worktree: none — design-only phase (deliverable = design note + locked API sketch on the issue; no source edits)
• repos:
• notes: Phase 1/4 of feature/autolens/multi_shared_state_examples (parent + phase-1 prompts in issued/; phases 2-4 prompts staged in feature/autolens/, deliberately un-issued). Claims no repos, so it does not conflict with kernel-forward-chunking / rect-adapt (PyAutoArray) or markdown-example-renderings (autolens_workspace); those claims matter from Phase 2 on.
• status-update 2026-07-10: DESIGN DELIVERED — decisions D1-D6 posted as the design-note comment on #599 (shared object = source-plane mesh geometry + H on AbstractPreloads; same-lambda = shared-mesh-per-factor now, joint stacked inversion as explicit follow-up prompt multi_joint_stack_inversion; shifts via existing aa.DatasetModel, known-by-default). Phase 2-3 prompts updated with the locked decisions. AWAITING user review of D2 especially ("one source" = per-factor-on-identical-mesh now, literal joint solve deferred) before issuing Phase 2 and closing #599's phase-1 leg.

autofit-assistant-birth

• issue: PyAutoLabs/autofit_assistant#1
• session: (this CLI session, 2026-07-10)
• status: workspace-dev — Phase 0 SHIPPED to main (3667f84); Phase 1 BUILT (d388ce1 local, 50/50 tests, audit/citations/idioms clean) PARKED awaiting-input at ship sign-off; Phase 2 (stats wiki) continuing on stacked branch
• question: PyAutoLabs/autofit_assistant#1 (comment) (2 scope calls: benchmark harness->Phase 4; 5 skills deferred to 1b/3)
• autonomy: supervised effective (--auto launched 2026-07-10; header supervised = feature too-large cap; NO heart-ack given — any Heart YELLOW at ship parks)
• launch-gates: answered in-session 2026-07-10 — repo PUBLIC at birth (dedicated repo-creation question); Phase 0 HAND-AUTHORED (clone-v1 plan as file-inventory checklist only); demos = 1D Gaussian + SNe cosmology distance-modulus fit
• note: Feature-Agent research re-home OVERRULED at launch (scoping done at intake; design in prompt); absorbs research/autofit_assistant/autofit_assistant_planning.md (retire at ship); PyAutoMemory never referenced in public output (Phase 2 pages are generalised rewrites)
• worktree: none (in-place: ~/Code/PyAutoLabs/autofit_assistant, new clone)
• repos:
  • autofit_assistant: main (phase 0), then feature/afa-phase-

multi-shared-state-examples-phase-3-workspace-examples

• issue: PyAutoLabs/autolens_workspace#260
• session: (this CLI session, 2026-07-10)
• status: workspace-dev
• worktree: ~/Code/PyAutoLabs-wt/multi-shared-state-phase-3
• repos:
  • autolens_workspace
  • autolens_workspace_test
  • autolens_profiling
• autonomy: safe effective (feature/medium, --auto granted by user 2026-07-10; plan on issue #260; four-leg gate; ends at PR-open, merge human)
• conflict-override: autolens_workspace claim markdown-example-renderings touches only markdown/ renders (0 files under scripts/multi/) — zero overlap, proceeding in fresh worktree
• epic: phase 3/4 of multi_shared_state_examples (phase 2 merged 2026-07-10: PyAutoArray#380 + PyAutoLens#600)