prompt: ship smoke-runner-delegation (#260) → complete

active.md

@@ -13,52 +13,3 @@
 - registered: 2026-08-19 by the wake_up session — the issuing session (claude/autofit-priors-messages-audit-ylvenv)
   filed the prompt + issue but not this entry, tripping Lifecycle Drift on main.
 - repos-none-claimed: this entry claims NO repos — one line deliberately, not 2-space bullets.
-
-## smoke-runner-delegation
-- issue: https://github.com/PyAutoLabs/PyAutoHands/issues/260 (issued 2026-08-24)
-- issued: 2026-08-24
-- prompt: active/run_smoke_copy_drift.md
-- status: library-dev
-- worktree: ~/Code/PyAutoLabs-wt/smoke-runner-delegation
-- decision: the prompt's blocking question is ANSWERED before any repo is touched —
-  the timeout/kill behaviour was PROMOTED to everyone (PyAutoHands#226/#227 → 52408a84;
-  all 10 copies import timeout_for + kill_group). Shape chosen: full delegation to a
-  PyAutoHands-owned runner, staged. Phase 1 = --list allowlist mode + the 4 *_workspace_test
-  copies; phase 2 = notebook leg + the 3 workspace copies; phase 3 = HowTo, no work.
-- measured: 2026-08-24, from every repo's main — 10 copies, 3 variants, ZERO functional
-  drift inside any variant (workspace 356L byte-identical; workspace_test 198L comment-only;
-  HowTo 75L differ by PROJECT alone). Prompt steps 1 and 4 verified already done.
-- registered: 2026-08-24 by the start_dev session (claude/smoke-copy-drift-ci-docs-ozntvv);
-  worktree_check_conflict clean (exit 0) for all 8 repos.
-- status-detail: HANDS LEGS MERGED, workspace collapses not started. The task stays
-  active deliberately — issue #260 scopes seven workspace PRs that have zero commits,
-  so this is not complete and must not be recorded as such.
-- library-pr: https://github.com/PyAutoLabs/PyAutoHands/pull/261 MERGED 2026-08-24 → c0e2e53
-  (both Hands legs: phase 1 `--list` opt-in script lists on run_python.py; phase 2
-  the matching notebook leg on run.py — `--list`, `--no-write-back`, `--retry-from`).
-  16 new tests; suite 388 passed / 5 skipped / 0 failed; CI green on all three
-  matrix legs (3.12/3.13/3.14) plus the tenant-firewall step in each.
-- mind-pr: https://github.com/PyAutoLabs/PyAutoMind/pull/280 MERGED 2026-08-24 → 0cd8544a
-- phase-2 findings (these changed the plan, recorded so the workspace PRs inherit them):
-  - run_notebook.py writes executed outputs back IN PLACE. Correct for generation
-    (the outputs are the product), wrong for a PR gate, which must not dirty the
-    tree it tests. Hence `--no-write-back`; it supersedes the workspace copy's
-    staged-copy-at-root trick rather than porting it, since the kernel cwd is
-    already pinned to the repo root.
-  - JUPYTER_MISSING_RC did NOT need promoting. It exists because the workspace copy
-    shelled out to a bare `jupyter`; execute_notebook invokes
-    `sys.executable run_notebook.py`, so the abort-with-no-summary failure mode is
-    structurally absent. One promotion item dissolved on inspection.
-- remaining: the 7 workspace collapses (4 *_workspace_test 198L → ~75L, then 3 workspace
-  356L → ~75L), each a one-file PR modelled on HowToLens's delegator. UNBLOCKED as of
-  2026-08-24 — the library-first gate cleared when PyAutoHands#261 merged. Phase 3
-  (HowTo ×3) needs no work; it is already the end state.
-  The delegator invocations these PRs should use:
-    *_workspace_test: run_python.py <project> scripts --list smoke_tests.txt
-                      --report-dir test-results
-    workspace:        the above, plus run.py <project> notebooks
-                      --list smoke_notebooks.txt --no-write-back --retry-from scripts
-  `--report-dir` is load-bearing: without it run_python.py never propagates failure
-  and the gate is vacuously green.
-- repos:
-  - PyAutoHands (branch claude/smoke-copy-drift-ci-docs-ozntvv)

active/run_smoke_copy_drift.md → complete/2026/08/smoke-runner-delegation.md

@@ -1,3 +1,118 @@
+Ten repos each carried their own `.github/scripts/run_smoke.py` — the file
+PyAutoHeart's reusable `smoke-tests.yml` invokes from inside each workspace, so
+one must exist per repo. What was duplicated was not the file's existence but
+its **driver loop**: read the allowlist, resolve per-entry env, spawn a
+subprocess under the timeout cap, classify PASS/FAIL/TIMEOUT, accumulate, print
+the summary, exit non-zero. Seven repos held that loop; the three HowTo repos
+were already thin shims.
+
+All ten are now shims. 2085 lines became 1127, and 1127 is mostly prose: of
+`autolens_workspace_test`'s 77 lines, 32 are the docstring explaining why it
+delegates.
+
+## The argument, which was not "they have drifted"
+
+The prompt was filed on a drift finding. Re-measured 2026-08-24 from every
+repo's `main`, there was **zero functional drift inside any variant** — the
+three workspace copies byte-identical, the four `_test` copies differing in two
+docstring lines, the three HowTo copies in `PROJECT` alone.
+
+That is not evidence the design was safe. It is the receipt for three manual
+N-repo sweeps that had been paid to restore it: the env-resolver fork (#185),
+the per-script timeout and process-group kill (#226/#227), and the jupyter
+guard. The HowTo tier needed **none** of the three, precisely because it holds
+no logic. The case for collapsing is maintenance cost, not drift.
+
+The prompt's blocking question — "does autolens_workspace_test's timeout/kill
+get promoted, or keep a documented divergence?" — was already answered by
+promotion in #226/#227. All ten copies imported `timeout_for` and `kill_group`
+before this task started; `_kill_group` existed in zero.
+
+## The real obstacle was discovery model, not behaviour
+
+The shared runners were opt-out (recursive discovery minus `no_run.yaml`); the
+seven logic-bearing copies are opt-in allowlists (`smoke_tests.txt`,
+`smoke_notebooks.txt`). That mismatch is why they could not delegate. So each
+phase was a PyAutoHands feature first and per-repo edits second.
+
+## Two bugs found by measuring before writing
+
+**1. The `no_run`-wins rule would have deleted 13 scripts from smoke coverage
+(#262).** #261 shipped "an explicit exclusion is the more specific intent, so
+`no_run` wins over the allowlist". Measured against the real repos, that rule
+would have silently skipped 9 scripts in autogalaxy_workspace_test, 2 in
+autolens_workspace_test, 1 in autofit_workspace and 1 in autolens_workspace —
+every one of which runs in smoke today, since the vendored runners never opened
+`no_run.yaml` at all.
+
+The rule was wrong because it conflated two policies for two different runs:
+`no_run.yaml` governs the release mega-run and notebook generation, the
+allowlist governs the PR gate, and a script legitimately appears in both. The
+failure mode is the dangerous kind — **CI would have stayed green**, because a
+skipped script is not a failure. Corrected to allowlist-authoritative before any
+workspace was touched, and confirmed in production afterwards: autogalaxy's nine
+all ran and passed, as did autolens's two and autofit's `searches/mcmc` in both
+its script and notebook forms.
+
+Same PR fixed a second blocker: `autocti_workspace_test` is the only workspace
+with no `config/build/no_run.yaml`, and the autohands-level fallback path does
+not exist either, so both runners crashed with `FileNotFoundError` before
+running anything.
+
+**2. `regenerate_notebook` resolved the source by bare filename (#263).** It
+looked up `scripts_dir / nb_path.name`, dropping the subdirectory, so
+`notebooks/imaging/model_fit.ipynb` searched for `scripts/model_fit.py`. Every
+workspace notebook is in a subdirectory, so the stale-notebook recovery was dead
+on arrival — and a bare filename can find the *wrong* source when two topics
+share a name. The existing tests missed it because their fixture was **flat**,
+which is the one layout where the bug is invisible. The replacement tests use a
+nested notebook plus a decoy script at the scripts root, and were negative-tested
+against the old resolution.
+
+## What was proven rather than assumed
+
+- **Env resolution is identical.** The old runners passed a *relative* script
+  path to `build_env_for_script`; the shared runners pass an *absolute* one. All
+  128 listed scripts across all seven repos were resolved both ways and diffed:
+  0 differences, in env and in args. Pattern matching is substring/stem based so
+  path form is irrelevant, and no profile pattern collides with the absolute
+  prefix.
+- No profile sets per-script `args`, so the shared runners' extra-args support
+  changes nothing.
+- Every allowlist entry resolves to a real file, and every listed notebook's
+  source script exists for the retry path.
+- The two-leg exit code is the worst of both, verified in all three
+  pass/fail combinations — a failing notebook cannot be masked by passing
+  scripts.
+- **Every merge was verified from the CI log, not the green tick**, because a
+  coverage regression here is indistinguishable from a pass. Each repo's
+  "Running N listed scripts/notebooks" was checked against its allowlist:
+  11/11, 22/22, 35/35, 3/3, 8/8+2/2, 14/14+2/2, 35/35+2/2.
+
+## Traps for anyone touching this again
+
+- `--report-dir` is load-bearing. `run_python.py` only propagates failure when a
+  report was built; without it the gate runs to completion and always exits 0.
+- `run_notebook.py` writes executed outputs back **in place**. Correct for
+  generation, where the outputs are the product; wrong for a PR gate, which must
+  not dirty the tree it tests. Hence `--no-write-back`.
+- `JUPYTER_MISSING_RC` did **not** need promoting. It existed because the
+  workspace copies shelled out to a bare `jupyter`; the shared runner invokes
+  `sys.executable run_notebook.py`, so the abort-with-no-summary mode is
+  structurally absent. One planned promotion item dissolved on inspection.
+- A bare `off`/`on`/`yes` entry in `no_run.yaml` parses as a YAML **boolean** and
+  crashes `should_skip` with `TypeError: argument of type 'bool' is not
+  iterable`. Hit while building a fixture; not fixed, no repo currently has such
+  a script name.
+
+## Still open
+
+The merged `claude/smoke-copy-drift-ci-docs-ozntvv` branches across nine repos
+were not deleted — this session's git proxy refuses delete refspecs
+(`send-pack: unexpected disconnect`). They are all proven merged into `main`.
+
+## Original prompt
+
 # run_smoke.py: three runner variants across 10 repos, no sync mechanism
 
 Type: maintenance

complete/index.md

@@ -6,7 +6,7 @@ Token-light navigation over the finished-work records (schema:
 only then grep a dated bucket. Curators: edit the band between the CURATED
 markers; everything below GENERATED is rebuilt.
 
-1093 records across 7 buckets.
+1094 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -194,6 +194,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [small-datasets-loader-pixel-scales](2026/08/small-datasets-loader-pixel-scales.md)
 - [small-datasets-regime-stamp](2026/08/small-datasets-regime-stamp.md)
 - [small-datasets-rmtree-committed-data](2026/08/small-datasets-rmtree-committed-data.md)
+- [smoke-runner-delegation](2026/08/smoke-runner-delegation.md)
 - [smoke-runner-jupyter-guard](2026/08/smoke-runner-jupyter-guard.md)
 - [spawn-autonomy-log-generated](2026/08/spawn-autonomy-log-generated.md)
 - [spawn-drift-self-heal](2026/08/spawn-drift-self-heal.md)

dashboard.html

@@ -123,7 +123,7 @@
 </head>
 <body>
 <header class="hero"><span class="orb">📋</span><h1>PyAuto<b>Mind</b><span class="kind">Dashboard</span></h1><p class="tag">Intent. Priority. Flow.</p></header><p class="lede">Every task the Mind is holding. Tap a task's 📋 and its <code>/start_dev</code> command is on your clipboard — paste it into a Claude Code chat to route Claude straight to that task. <a href="#recent">Recent</a> is the same work by date — what has been happening rather than what to do next.</p>
-<ul class="stats"><li><b>2</b><span>In flight</span></li><li><b>3</b><span>Parked</span></li><li><b>6</b><span>Planned</span></li><li><b>155</b><span>Backlog</span></li></ul>
+<ul class="stats"><li><b>1</b><span>In flight</span></li><li><b>3</b><span>Parked</span></li><li><b>6</b><span>Planned</span></li><li><b>155</b><span>Backlog</span></li></ul>
 <p class="muted mdsrc"><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/dashboard.md">markdown version</a></p>
 <h2>Start here</h2>
 <h3>Highest priority <span class="facets">(filed as high) — showing 12 of 17</span></h3>
@@ -145,7 +145,6 @@ <h3>Quick wins <span class="facets">(small enough, and safe enough to run unatte
 <h2>In flight <a class="mdsrc" href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/active.md">markdown version</a></h2>
 <p class="muted">Issued — each has an open GitHub issue and usually a branch.</p>
 <div class="task"><button class="copy" data-cmd="/start_dev active/16_transformed_message_factor_gradient_unpack.md" aria-label="Copy the Claude command">📋</button><p><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/active/16_transformed_message_factor_gradient_unpack.md"><code>@PyAutoFit</code> <code>TransformedMessage.factor_gradient</code> crashes on first call</a> — <a href="https://github.com/PyAutoLabs/PyAutoFit/issues/1501">issue #1501</a><span class="facets"> — issued 2026-08-19</span> — <span class="facets">HOLD — do not start dev. Fix-or-delete hangs off the PyAutoFit#1498 logpdf-contract</span></p></div>
-<div class="task"><button class="copy" data-cmd="/start_dev active/run_smoke_copy_drift.md" aria-label="Copy the Claude command">📋</button><p><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/active/run_smoke_copy_drift.md">run_smoke.py: three runner variants across 10 repos, no sync mechanism</a> — <a href="https://github.com/PyAutoLabs/PyAutoHands/issues/260">issue #260</a><span class="facets"> — issued 2026-08-24</span> — <span class="facets">library-dev</span></p></div>
 <h2>Parked <a class="mdsrc" href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/parked.md">markdown version</a></h2>
 <details>
 <summary>3 task(s)</summary>
@@ -326,12 +325,6 @@ <h2>Backlog <a class="mdsrc" href="https://github.com/PyAutoLabs/PyAutoMind/tree
 <p class="muted">The 50 newest things to happen to the work in hand, newest first — issued, parked, filed. Every other section on this page is laid out by state, which is exactly why none of them can answer “what has been happening?”. Shipped work is not here: it is read from <code>complete/index.md</code>, and a thousand records deep it would crowd out everything anyone can still act on. Showing the newest 10; … opens the next 10.</p>
 <table class="recent">
 <tr>
-<td class="when">2026-08-24</td>
-<td class="what">issued</td>
-<td><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/active/run_smoke_copy_drift.md">run_smoke.py: three runner variants across 10 repos, no sync mechanism</a></td>
-<td class="pick"><button class="copy" data-cmd="/start_dev active/run_smoke_copy_drift.md" aria-label="Copy the Claude command">📋</button></td>
-</tr>
-<tr>
 <td class="when">2026-08-23</td>
 <td class="what">filed</td>
 <td><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/draft/maintenance/workspaces/pynufft_removal_downstream_residue.md">pynufft removal: unswept downstream residue (1 hard break + stale…</a></td>
@@ -385,7 +378,7 @@ <h2>Backlog <a class="mdsrc" href="https://github.com/PyAutoLabs/PyAutoMind/tree
 <td><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/draft/maintenance/libraries/remove_pynufft_legacy_transformer.md">Remove pynufft + legacy TransformerNUFFTPyNUFFT</a></td>
 <td class="pick"><button class="copy" data-cmd="/start_dev draft/maintenance/libraries/remove_pynufft_legacy_transformer.md" aria-label="Copy the Claude command">📋</button></td>
 </tr>
-<tr hidden>
+<tr>
 <td class="when">2026-08-22</td>
 <td class="what">filed</td>
 <td><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/draft/bug/pyautolens/point_source_json_datasets_record_no_regime.md">Point-source JSON datasets record no resolution regime</a></td>
@@ -625,6 +618,12 @@ <h2>Backlog <a class="mdsrc" href="https://github.com/PyAutoLabs/PyAutoMind/tree
 <td><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/draft/maintenance/autolens_workspace_developer/stale_api_rot_audit.md">autolens_workspace_developer: broad stale-API rot (56 symbols, no CI)</a></td>
 <td class="pick"><button class="copy" data-cmd="/start_dev draft/maintenance/autolens_workspace_developer/stale_api_rot_audit.md" aria-label="Copy the Claude command">📋</button></td>
 </tr>
+<tr hidden>
+<td class="when">2026-08-04</td>
+<td class="what">filed</td>
+<td><a href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/draft/bug/workspaces/aplt_output_drift_remaining_repos.md"><code>aplt.Output</code> stale-API drift in the remaining workspace repos</a></td>
+<td class="pick"><button class="copy" data-cmd="/start_dev draft/bug/workspaces/aplt_output_drift_remaining_repos.md" aria-label="Copy the Claude command">📋</button></td>
+</tr>
 </table>
 <button class="more" data-page="10">… 10 more (40 left)</button>
 <h2>Epics <a class="mdsrc" href="https://github.com/PyAutoLabs/PyAutoMind/blob/main/epics.md">markdown version</a></h2>