flow-next v1.10.0

Changed

• Eval-driven prompt optimization — 8 scout/analyst agents made ~40–71% leaner per call, with accuracy held (fn-54 / FLOW-5). Rolled the external "autoresearch" eval loop (baseline → one mutation → keep-if-better ratchet; methodology in agent_docs/optimizing-skills.md) across the read-only agents whose free-form output flows into the planner / work-loop context. Each gained a feature-preserving output budget — the reductions are at runtime (the rendered output), not in prompt size:
  • repo-scout (83→100% on its eval set, ~40–50% leaner) · context-scout (60→93%, ~60–70%, dropped the prescribed Code-Signatures block) · flow-gap-analyst (~50–70%, 26/27 gaps held) · quality-auditor (~63%) · spec-scout (No-Relationship → count, scale-robust) · docs-scout (~48–69%) · github-scout (~71%, the biggest) · practice-scout (~52%).
  • Feature-preservation is the guarantee, not a hope. Every mutation was kept only if a per-target coverage/accuracy eval held (the ratchet): grounding (context-scout cited paths test -f-verified vs ~/work/DocIQ-Sphere), findings (quality-auditor vs the ~/work/slop-testbed 7-issue corpus — Major bug + all slop still caught, clean stays ✅), gaps (per-input answer keys), and docs/APIs/gotchas (the "pointer-not-paste" rule: name the API inline, drop code blocks, the link carries depth). The leaner research scouts even surfaced extra real issues a verbose baseline missed (a current CVE; an extra trust-proxy gotcha).
  • End-to-end verified: the optimized scouts → a planner produced a correct, ship-quality build plan for a deliberately hard, cross-cutting DocIQ-Sphere feature (org-scoped agent-run rate limiting) reading only the budgeted scout output — features preserved at the consumer level, not just scout-output level.
• /flow-next:make-pr: removed stale fn-42.N build-scaffolding archaeology from the skill prompt (heading labels, a phase-reference table column, two orphaned sentences) — render output behaviorally identical; no guardrail / routing / tracker-sync logic touched.

Notes

• /flow-next:capture is unchanged. A trim was tried and reverted — it regressed business-context routing on one input (the ratchet caught it). The capture override guard (refuses to silently overwrite a user-edited spec) was verified intact.
• No flowctl / Python logic changed — prompt markdown only. Each agent edit was re-mirrored to its Codex copy via sync-codex.sh. Retained eval harnesses (frozen inputs, evals, baselines, per-experiment changelogs) under optimization/.