second-opinion-honesty v0.2.0 — opt-in enforced acknowledgement gate

Adds the optional enforced acknowledgement gate (做法 3) on top of v0.1.0.

When a project opts in (.second-opinion/require_ack), a project-scoped Stop-hook blocks the session from ending until the latest DRAFT run has a Second Opinion review and a human has acknowledged it.

Highlights

• Promotes the human-owned checkpoint from prose to a code-enforced gate (the iron rule: real enforcement lives in hooks, not prose).
• Enforces THAT a human acknowledged — never WHAT they decided. Anti-loop capped by SECONDOP_MAX_ACK_RETRIES (default 3); a silent no-op unless opted in, so a global install never blocks unrelated projects.
• New: secondop/gate.py (pure decision), secondop/ack.py (python3 -m secondop.ack — the human action; refuses to ack an unreviewed run), hooks/stop_ack.py + Stop wiring.
• 87 tests green (TDD; incl. a subprocess hook integration test); stdlib-only; CI green.

Enable (default-off, per project)

```
mkdir -p .second-opinion && touch .second-opinion/require_ack
```
Install/upgrade as in v0.1.0: /plugin install second-opinion-honesty@agentic-sop-to-work.

second-opinion-honesty v0.1.0 — domain-neutral honesty reviewer

First release of second-opinion-honesty — an independent, domain-neutral adversarial honesty reviewer that pairs with agentic-sop-kit. It checks whether a finished DRAFT's claims are backed by its evidence before a human approves (fabrication, transcription slips, invented IDs, wrong PASS/FAIL, conclusion overreach). It never approves or rejects — a human owns the verdict.

Highlights

• Deterministic engine (stdlib-only): #1 wrong verdict + aggregate recompute · #2 unsourced numbers (nearest-token transcription hint) · #3 invented identifiers/dates → 【待補】.
• FULL mode (reads a kit run_dir's trace, HARD @ 1.0) and DEGRADED mode (any document + supplied inputs, SOFT @ 0.5).
• Capped advisory LLM layer (#4 overreach): clamp to SOFT / conf ≤ 0.5, evidence-gate, settled-key suppression, SECONDOP_MAX_LLM_PASSES cap — all code-enforced.
• Report frozen to verdict=ADVISORY_ONLY + human_owns_verdict=true — it cannot say "approved".
• /second-opinion command + skill; cross-domain fixtures (pharma / FE / BE / plain doc); 69 tests; stdlib-only.

Install

```
/plugin marketplace add s0912758806p/agentic-sop-to-work
/plugin install second-opinion-honesty@agentic-sop-to-work
```
Pairs with — but installs independently of — agentic-sop-kit. Not a source-code reviewer (that's /code-review) and not a workflow-architecture auditor (that's agentic-workflow-audit).

v1.5.4 — flow GIF + smart intake (P1) + capped fix-loop (P2)

Added

• Smart intake for agentic-sop — classifies the input on trigger: a written spec / existing skill / filled SOP → use as-is; a natural-language need → Claude drafts a Human SOP (gaps 【待補】, no fabrication), pauses for human confirm, then continues to decomposition. (Generative layer only; gates + DRAFT+human spine unchanged.)
• Capped auto fix-loop — run.py emits a machine-readable failure{step,gate_type,message,artifact}, and /sop-flow auto-fixes & re-runs on a gate failure, capped by run.py --max-fix-retries (default 3, code-enforced per run-id → fix_exhausted); exhausted → stop for a human; never patches output to pass. Shares one cap knob SOPKIT_MAX_FIX_RETRIES (default 3) with the Stop-hook regression loop — two layers, one setting. Adds tests/integration/test_fix_loop.py.

Changed

• README "At a glance" is now an animated step-by-step flow GIF (assets/flow.gif, generated by assets/flow/make_flow_gif.py — Pillow, deterministic); the static Mermaid diagram moved into a <details> fallback. Depicts intake routing + the capped fix-loop.

Full changelog: https://github.com/s0912758806p/agentic-sop-to-work/blob/main/CHANGELOG.md

v1.5.3 — CSO Use-when skill descriptions

v1.5.3 — CSO Use-when skill descriptions

Skill discoverability fix. Rewrote the agentic-sop and agentic-workflow-audit skill descriptions to CSO (Claude-Search-Optimization) form so they auto-trigger reliably.

Changed

• agentic-sop — description 642 → 331 chars
• agentic-workflow-audit — description 587 → 379 chars
• Both now lead with Use when… (triggering conditions only), drop the embedded process summary (three/seven-phase, gate names), and add reciprocal 不適用 disambiguation so build requests route to agentic-sop and audit requests to agentic-workflow-audit.
• Added CHANGELOG.md (full history 1.0.0 → 1.5.3).

Why

The old descriptions led with what the skill does and embedded a workflow summary, which (a) violates CSO — a description should be triggering conditions only — and (b) added to the skill-catalog description-budget pressure that was dropping these descriptions from the live catalog entirely. Shorter, trigger-first descriptions restore catalog rendering and improve auto-trigger; once a description renders, triggering no longer depends on a project CLAUDE.md.

Validation

• A/B routing trials (oblique + English prompts): 100% correct routing, no regression vs the old descriptions.
• Regression suite: 14/14 pass. Frontmatter YAML round-trips exactly.

Update

/plugin marketplace update

then update agentic-sop-kit to v1.5.3.

Full changelog: https://github.com/s0912758806p/agentic-sop-to-work/blob/main/CHANGELOG.md

v1.5.2 — engine extraction + bilingual errors

v1.5.2 — engine extraction + bilingual errors

Changed

• Extracted the SOP orchestration engine (run_step / run_map / print_plan) from run.py into kit/lib/engine.py; slimmed run.py to orchestrator-only.
• Bilingual (EN / 中文) engine error messages; expanded --plan documentation.

Full changelog: https://github.com/s0912758806p/agentic-sop-to-work/blob/main/CHANGELOG.md

v1.5.1 — --plan branch/map + static goto validation

v1.5.1 — --plan branch/map rendering + static goto validation

Fixed

• --plan now renders branch / map steps and statically validates branch goto targets (forward-only, must exist, no duplicates).
• Duplicate step-name flagged only at the goto site; malformed steps flagged; missing goto handled.
• Added backward-goto and duplicate-name tests.

Full changelog: https://github.com/s0912758806p/agentic-sop-to-work/blob/main/CHANGELOG.md