Resolves #13956
Bounds the Sandman handoff output where it was expanding beyond its boot-context role: Recent Open PRs now defaults to 10 and renders an (x / y) count, the redundant PRs-by-author expansion is removed, topology conflict alerts are rewritten through a 5-item cap, and Computed Golden Path keeps Strategic Interpretation present with the model-generated brief when valid plus a deterministic fallback when the model path is unavailable.
Evidence: L2 unit coverage satisfies the #13956 ACs; no L3/L4 evidence is required because the changed behavior is pure/local handoff rendering.
Deltas From Ticket
- Updated
learn/agentos/DreamPipeline.md and learn/agentos/wake-substrate/sandman-handoff-format.md so the documented handoff shape matches the compact renderer and the new Strategic Interpretation fallback.
- Ran the repo block-alignment fixer on the touched GoldenPathSynthesizer spec after the pre-commit hook flagged new alignment drift.
Config Template Change
- Changed key:
goldenPathRecentOpenPrRenderLimit
- Default changed from
5 to 10.
- Local
config.mjs files do not need manual shape/key updates unless they explicitly pinned the old value and want the new default.
- Harness/orchestrator restart is recommended after merge for daemons to pick up the new default.
Substrate Slot Rationale
Modified reference docs under learn/agentos/**; disposition keep. The docs now prevent future sessions from rebuilding removed boot-context bloat and clarify that Strategic Interpretation must remain present via fallback.
Test Evidence
git diff --cached --checknpm run test-unit -- test/playwright/unit/ai/services/graph/GoldenPathSynthesizer.spec.mjs test/playwright/unit/ai/services/rem-observability.spec.mjs → 58 passed- Pre-commit hook passed: whitespace, shorthand, AiConfig mutation lint, JSDoc type lint, ticket archaeology, block alignment
Post-Merge Validation
Commit
56f6f6e085 — fix(ai): bound Sandman handoff output (#13956)
Authored by Euclid (GPT-5, Codex Desktop). Session cd4ef6e5-e52d-475f-a8b1-a9b781c5c91d.
Resolves #13956
Bounds the Sandman handoff output where it was expanding beyond its boot-context role: Recent Open PRs now defaults to 10 and renders an
(x / y)count, the redundant PRs-by-author expansion is removed, topology conflict alerts are rewritten through a 5-item cap, and Computed Golden Path keepsStrategic Interpretationpresent with the model-generated brief when valid plus a deterministic fallback when the model path is unavailable.Evidence: L2 unit coverage satisfies the #13956 ACs; no L3/L4 evidence is required because the changed behavior is pure/local handoff rendering.
Deltas From Ticket
learn/agentos/DreamPipeline.mdandlearn/agentos/wake-substrate/sandman-handoff-format.mdso the documented handoff shape matches the compact renderer and the new Strategic Interpretation fallback.Config Template Change
goldenPathRecentOpenPrRenderLimit5to10.config.mjsfiles do not need manual shape/key updates unless they explicitly pinned the old value and want the new default.Substrate Slot Rationale
Modified reference docs under
learn/agentos/**; dispositionkeep. The docs now prevent future sessions from rebuilding removed boot-context bloat and clarify that Strategic Interpretation must remain present via fallback.Test Evidence
git diff --cached --checknpm run test-unit -- test/playwright/unit/ai/services/graph/GoldenPathSynthesizer.spec.mjs test/playwright/unit/ai/services/rem-observability.spec.mjs→ 58 passedPost-Merge Validation
resources/content/sandman_handoff.md; verify Recent Open PRs is capped/count-rendered, no### @neo-*PR author sections appear, conflict alerts stay at 5, and Strategic Interpretation is present.Commit
56f6f6e085—fix(ai): bound Sandman handoff output (#13956)Authored by Euclid (GPT-5, Codex Desktop). Session cd4ef6e5-e52d-475f-a8b1-a9b781c5c91d.