You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Source: P12 adversarial evaluation, finding F6 · MED — docs.local/plan/stability-v2/phase-12/evaluation.md:129-140
Evaluated against the deployed build: 0.4.46, HEAD 4a88b43. P11 shipped in #454 / v0.4.45.
This is the tracking issue for the P11b work that until now existed only as prose.
Evidence (verbatim from the evaluation, evaluation.md:129-140)
F6 · MED · P11's delivery half is unwired, untracked, and unreachable on resume
coordination_footer_delivered = false is hardcoded (server.ts:11820). Honestly disclosed in three
places, and the reasoning (618 chars vs a 500-char threshold moving every spawn onto the incident-prone
chunked path) is sound engineering judgment. But:
spawn(resume_agent_id) returns no report_path/done_marker at all (verified live), so a lead
resuming a worker cannot relay the contract it is required to relay.
U10's goal — "briefs/reports stop being ad-hoc lead inventions" — is therefore unmet by design today. Self-demonstrating: my own contract was hand-relayed by the lead, and my probe reported back
unprompted: "the inbox was empty, so the relayed REPORT_PATH/DONE_MARKER never arrived."
Scope of P11b
File-carried contracts — wire the delivery half so the engine-issued report_path/done_marker actually reach the spawned agent, without pushing every spawn onto the incident-prone chunked path (the 618-chars-vs-500-threshold constraint that motivated hardcoding coordination_footer_delivered = false).
Resume parity — spawn(resume_agent_id) must return report_path/done_marker. Today it returns neither (verified live), so a lead resuming a worker cannot relay the contract it is required to relay.
Until both land, U10's goal — "briefs/reports stop being ad-hoc lead inventions" — is unmet by design.
Source: P12 adversarial evaluation, finding F6 · MED —
docs.local/plan/stability-v2/phase-12/evaluation.md:129-140Evaluated against the deployed build: 0.4.46, HEAD
4a88b43. P11 shipped in #454 / v0.4.45.This is the tracking issue for the P11b work that until now existed only as prose.
Evidence (verbatim from the evaluation, evaluation.md:129-140)
Scope of P11b
report_path/done_markeractually reach the spawned agent, without pushing every spawn onto the incident-prone chunked path (the 618-chars-vs-500-threshold constraint that motivated hardcodingcoordination_footer_delivered = false).spawn(resume_agent_id)must returnreport_path/done_marker. Today it returns neither (verified live), so a lead resuming a worker cannot relay the contract it is required to relay.Until both land, U10's goal — "briefs/reports stop being ad-hoc lead inventions" — is unmet by design.
Evaluation:
docs.local/plan/stability-v2/phase-12/evaluation.md— cmuxlayer-issue-filer (worker) · claude-code/claude-opus-5