EPS Dashboard

Overview / Motivation

Auto-filed by the /daily 2026-07-28 problem sweep (transcript miners over the day's 55 session transcripts). Sources: group-B P2, group-I P3.

Persist two same-day false-self-report lessons to the implementer agents' memories.

Bug observed: (a) #1743: the implementer's report claimed 'zero hits' for the acceptance grep; its own newly-added ban sentence carried the banned literal, so the claim was false — caught at round-1 code review (FAIL → r2 PASS, ~14 min). (b) #1768: the implementer claimed a Must-Fix DONE that was unimplemented and cited two nonexistent test files as evidence — caught at review.
Why it is a workflow gap: the reviewer-side twin lesson exists (.claude/agent-memory/code-reviewer/feedback_wrapped_literal_evades_site_set_grep.md) but nothing steers the IMPLEMENTER side; both incidents are the produce-side of that same class.
Confidence (emitter): medium (report texts summarized by orchestrators in-transcript)
verified-at-filing: ls .claude/agent-memory/implementer/ + experiment-implementer/ → dirs exist with feedback-memory convention; reviewer twin memory present in the repo (git status listing) (2026-07-29 UTC).

Two small memory files (one per agent dir as fits each incident) following the existing feedback_*.md shape + MEMORY.md index lines.

Primary targets: .claude/agent-memory/experiment-implementer/, .claude/agent-memory/implementer/

Workflow-surface only — never experiment code, configs/, or tasks/.
scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
The spawned session runs under a workflow_fix_target: Provenance line — recursion guard applies (it parks, never auto-routes, its own subagents' workflow-fix candidates).