fix(refinery): include gc.routed_to=polecat in prompt Rejection Flow#11
Open
eric-jones wants to merge 1 commit into
Open
fix(refinery): include gc.routed_to=polecat in prompt Rejection Flow#11eric-jones wants to merge 1 commit into
eric-jones wants to merge 1 commit into
Conversation
eric-jones
force-pushed
the
topic/refinery-rejection-routing
branch
from
1752368 to
e5edc5b
Compare
eric-jones
force-pushed
the
topic/refinery-rejection-routing
branch
from
e5edc5b to
7282cec
Compare
julianknutsen
added a commit
that referenced
this pull request
May 21, 2026
…r 3 (gastownhall#1594) ## Summary Make the equivalence between the worker's claim path (`EffectiveWorkQuery` Tier 3) and the reconciler's spawn path (`EffectiveScaleCheck` / new `EffectivePoolDemandQuery`) **structural** rather than coincidental. PR gastownhall#1516 symmetrically simplified both predicates to "ready unassigned routed_to" by retiring the molecule tier from each side. Post-gastownhall#1516 the two are textually equivalent — but they remained two separate code paths. A future PR adding a tier to one without the other would silently re-introduce the protocol-mismatch class addressed by fo-spawn-storm (the reconciler spawning sessions for work the worker can't claim, or vice versa). This refactor closes the residual gap by extracting the shared predicate into a single helper: ```go func bdReadyPoolDemandShell(target string) string { return `bd ready --metadata-field gc.routed_to=` + target + ` --unassigned --json` } ``` Both `EffectiveWorkQuery` Tier 3 (with `--limit=1` appended, first-row form) and the new `EffectivePoolDemandQuery` (piped through `jq 'length'`, count form) derive from this helper. The reconciler's `scaleParamsFor` now calls `EffectivePoolDemandQuery` directly so the dependency on the work-query predicate is explicit at the call site. `EffectiveScaleCheck` remains as a thin pass-through for back-compat with code and configs that name the predicate "scale_check". ## Tests - **Structural** — `TestPoolDemandPredicateSharedWithWorkQuery` asserts both methods reference the `bdReadyPoolDemandShell` output for the same target. Future divergence becomes a compile-touched test failure rather than a quiet behavior regression. - **Behavioral** — `TestPoolDemandAndWorkQueryAgreeOnRoutedSemantics` runs both queries against the same fake `bd` and asserts they agree on \"routed work present\" / \"no routed work\". This is the behavioral counterpart called out in fo-glz1h: a state the worker can't claim must produce demand=0 from the reconciler. - **Override pass-through** — `TestEffectivePoolDemandQueryRespectsOverride` confirms a user-set `scale_check` short-circuits both methods. ## Docs `engdocs/architecture/dispatch.md` gets: - A new \"scale_check ↔ work_query correspondence\" section explaining the two read sides and why they must stay symmetric, with the spawn-storm pre-gastownhall#1516 symptom called out as the failure mode. - A new invariant (#11) that any tier-shape change to one MUST flow through `bdReadyPoolDemandShell` to the other, with the regression tests named. - The Code Map row for `internal/config/config.go` gains the new symbols. - \"Last verified against code\" date bumped to 2026-05-01. ## Stack Based on `fix/main-auto-port-bind-retry-20260501` (PR gastownhall#1582) so the unrelated tempfile flake on `main` doesn't keep failing CI for this branch. ## Test plan - [x] `go test ./internal/config/...` passes (existing + 3 new tests). - [x] `go vet ./...` clean. - [x] Targeted pool tests (`TestEvaluatePool*`, `TestScaleParams*`, `TestPool*`) pass. - [ ] Full CI run on the PR (broader cmd/gc suite has pre-existing failures on `fix/main-auto-port-bind-retry-20260501` — TestCmdHookPoolInstanceUsesTemplatePoolLabel, TestDoPrimePoolAgentFallback, TestControllerStateCreateRigPokesReconciler, TestReconcileCitiesNameDriftStopsBeadsProvider — verified to fail on this base before any of these changes; not introduced by this PR). ## Related - Closes fo-glz1h (surviving \"Keep\" item from the now-closed fo-78wj3, plus the dispatch.md updates flagged by gc-wisp-acvu). - Follows up on PR gastownhall#1516 (commit 6b5d912) — the symmetric simplification this builds on. - Closes the residual structural gap behind fo-spawn-storm (DEFERRED). <!-- codesmith:footer --> --- <a href="https://app.blacksmith.sh/gastownhall/codesmith/gascity/pr/1594"><picture><source media="(prefers-color-scheme: dark)" srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"><source media="(prefers-color-scheme: light)" srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-light.svg"><img alt="View in Codesmith" src="https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"></picture></a> <sup>Need help on this PR? Tag <code>@codesmith</code> with what you need.</sup> - [ ] Let Codesmith autofix CI failures and bot reviews <!-- /codesmith:footer --> --------- Co-authored-by: Julian Knutsen <julianknutsen@users.noreply.github.com>
![https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"><source](https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"><source){kind=link}
![https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-light.svg"><img](https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-light.svg"><img){kind=link}
![https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"></picture></a](https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"></picture></a){kind=link}
eric-jones
force-pushed
the
topic/refinery-rejection-routing
branch
2 times, most recently
from
0d1d80c to
b39e555
Compare
The refinery prompt's "Rejection Flow" quick reference showed a partial `gc bd update $WORK --status=open --assignee="" --set-metadata rejection_reason="..."` that omitted `--set-metadata gc.routed_to=...polecat`. LLM agents ran exactly that partial form (verbatim from the prompt) instead of the formula's full update. Result: rejected work beads sat in the queue with stale `gc.routed_to=...refinery` (set when the polecat originally submitted to refinery), and the pool reconciler — which matches on `gc.routed_to`, not `rejection_reason` — never spawned a polecat to resume. The bead stayed open forever; the convoy stalled. Reproduced on validate-l5-20260510-v4 (gascity local/integration 4d656c14 with gastownhall#1925, gastownhall#1916, gastownhall#1913 in flight): two L5c work beads with conflicting edits. Refinery merged the first as a fast-forward and rejected the second correctly, but the rejected bead still had `gc.routed_to=...refinery` after the rejection update — verified directly from the rig's `.beads/issues.jsonl` and the refinery agent's transcript. The agent's exact rejection command: gc bd update vl2vr-j4y --status=open --assignee="" \ --set-metadata rejection_reason="rebase conflict on shared_func.go: ..." — matches the prompt's incomplete flow, not the formula's full one. ## What changed - `examples/gastown/packs/gastown/agents/refinery/prompt.template.md` — rewrote "Rejection Flow" to show the full bd update with `--set-metadata gc.routed_to="${GC_RIG:+$GC_RIG/}{{ .BindingPrefix }}polecat"`, plus a one-paragraph "why" naming the pool reconciler match. Pointed readers at the formula's `rebase` / `handle-failures` step as canonical. - `examples/gastown/gastown_test.go` — `TestRefineryPromptRejectionFlowIncludesRoutedTo` asserts the prompt's Rejection Flow contains both `--set-metadata rejection_reason=` and the full `gc.routed_to=...polecat` line, in order. Guards against silently reverting to a partial bd update. Surface: refinery rejection metadata via `gc bd update --set-metadata` (canonical bd CLI), shown in canonical agent prompt template (rendered by `gc init`); not direct Dolt edits, not inline ad-hoc metadata queries. Fixes: stg-66wz (Issue 2: spawn-on-rejection) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Upstream-PR: gastownhall#1934
eric-jones
force-pushed
the
topic/refinery-rejection-routing
branch
from
b39e555 to
9a239af
Compare
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.
Fork-PR for review. Upstream: gastownhall#1934 (OPEN). julianknutsen asked for manual verification; gc-rm0ha.49 schedules L5c rerun.