Blockers First Governance Hub

The MANDATORY Blockers First protocol (CLAUDE.md §“MANDATORY: Blockers First”) requires that before finalizing any implementation steps, Claude must surface all unknowns as numbered questions (Q1…Qn), each with answer options, recommendation, explanation, and “if Henry picks otherwise” delta. Every question must carry a recommendation — bare questions without a recommendation are a governance violation. This hub documents the Recommendation+Explanation pattern, the G-BLOCKER-SURFACING gate, and all related feedback rules. Enforcement mode is manual — Claude self-enforces at plan-authoring time; the G-BLOCKER-SURFACING gate additionally requires mirroring open blockers into SESSION-AUDIT.md.

Quick reference

Field	Value
Gate ID	G-BLOCKER-SURFACING (§A1, gate #3)
CLAUDE.md section	§“MANDATORY: Blockers First” + §A1 G-BLOCKER-SURFACING
Enforcement mode	manual (self-enforced at plan-authoring; G-BLOCKER-SURFACING adds SESSION-AUDIT mirror)
Validator script	n/a
Discord alert	pending (Phase 2 weekly-index-audit.timer)
Feedback rules	feedback_blocker_surfacing, feedback_question_discipline, feedback_options_listed_vertically, feedback_time_boxed_confirm, feedback_scope_echo_required
Incidents derived from	47 hidden blockers across 32 plans (2026-05-01 audit); cloudflare Q1-Q3 open 11 days; salesmsg-ctie Q6 gating Phases 3-9; Aurora SOUL overwrite attempt
Violable by	Claude
Last known violation	2026-05-01: 47 hidden blockers discovered across 32 plans — none mirrored to SESSION-AUDIT NEXT ACTIONS
Amendment	§A1 2026-05-01 (G-BLOCKER-SURFACING gate formally added)
Phase status	active
Last audit	2026-05-03

Components

• ~/CLAUDE.md §“MANDATORY: Blockers First” — authoritative protocol text
• ~/CLAUDE.md §A1, gate #3 (G-BLOCKER-SURFACING) — enforcement rule
• /home/opsadmin/.openclaw/workspace/SESSION-AUDIT.md — mirror target for unresolved blockers
• /home/opsadmin/.openclaw/workspace/HENRY-PREFERENCES.md §4 — “Every Question Must Carry a Recommendation”
• /home/opsadmin/.claude/projects/-home-opsadmin/memory/feedback_blocker_surfacing.md — incident-derived rule
• /home/opsadmin/.claude/projects/-home-opsadmin/memory/feedback_question_discipline.md — question formatting rule

Recommendation+Explanation pattern (mandatory per question)

Every Q1…Qn in any plan or interaction MUST include all four parts:

**Q<N>: <question title>**

**Options:** A) <option> B) <option> C) <option>
**Recommend:** <option letter + name — stated explicitly>
**Why:** <tradeoff / leverage / risk / constraint driving the call>
**If Henry picks otherwise:** <what changes — cost of overriding>

Bare questions without a recommendation are a governance violation. Henry is the decider, not the analyst. Make the call, show the work, let him override.

How it’s used

• Trigger condition: Before finalizing any implementation plan that touches external APIs, modifies services, or creates multi-phase work
• Workflow: (1) List unknowns as Q1…Qn → (2) For each: state Options + Recommend + Why + Override cost → (3) If proceeding with assumptions, mark clearly as ASSUMPTION → (4) Mirror open blockers to SESSION-AUDIT.md NEXT ACTIONS → (5) Never finalize plan with unresolved blockers
• Agents involved: Not agent-invoked — applies to Claude-as-planner in main conversation
• Failure mode: Hidden blockers: plan proceeds, implementation hits wall mid-phase → rollback cost, rework, wasted Henry time
• Success criteria: Every plan has explicit Q&A section; no ## Open Blockers items absent from SESSION-AUDIT NEXT ACTIONS within same session

G-BLOCKER-SURFACING rule (CLAUDE.md §A1 gate #3)

Any plan with ## Open Blockers / ## Decision Gates / ## Open Questions containing unresolved items MUST mirror those into SESSION-AUDIT.md NEXT ACTIONS or OPEN ISSUES within the same session the blocker is identified.

What this stops: The 47 hidden blockers across 32 plans that were silently dormant as of 2026-05-01 audit. Specific examples:

• cloudflare Q1-Q3: 11 days open without surfacing
• salesmsg-ctie Q6: 5-min Henry task gating Phases 3-9 entire branch
• Aurora SOUL overwrite: would have overwritten “Build, don’t coordinate” principle — caught because blocker was raised pre-execution

HENRY-PREFERENCES.md §4 integration

From workspace/HENRY-PREFERENCES.md §4 “Every Question Must Carry a Recommendation”:

Henry is the decider, not the analyst. Claude must make the call, show the work, and let Henry override. A question without a recommendation is forcing Henry to do Claude’s job.

This means: even when uncertain, Claude must pick a recommended option with a stated reason. Saying “it depends” without a recommendation is a governance violation.

Cross-links

Agents that touch this

• _summary — Aurora raises blockers before outreach campaigns
• _summary — acquisition plans frequently have open Q&A sections
• _summary — backfill jobs raise blockers on schema changes

Skills that invoke this

• acquisitions-outreach — blockers surface as B1-B16 in OSIL ratification
• hubspot-deal-ingest — dedup + dual-pipeline blockers surfaced before write

Plans that govern this

• openclaw-fragmentation-fix-2026-05-01 — §A1 gate #3; Phase 2 enforcement crons
• openclaw-self-improvement-layer-2026-05-03 — B1-B16 pattern of structured blockers
• nemoclaw-audit-2026-05-03 — B1-B6 structured with full R+E pattern

Feedback rules

• feedback_blocker_surfacing — G-BLOCKER-SURFACING incident-derived rule
• feedback_question_discipline — question formatting + mandatory recommendation
• feedback_options_listed_vertically — A/B/C format for option presentation
• feedback_time_boxed_confirm — blockers need time-box or they become infinite
• feedback_scope_echo_required — echo what’s in scope before acting
• feedback_no_partial_completion — blocked work must be surfaced, not silently dropped
• feedback_post_audit_persistence — blockers found in audit stay in audit until resolved

KB / source docs

• README — system-level planning conventions

System maps

• governance-gates-network — G-BLOCKER-SURFACING in gate topology

Related cluster

Governance enforcement cluster (anchor: g-gates-network):

• sources-first — sources verified before Q surfaced
• action-gate — action gate confirms after Q resolved
• plan-governance — plan Q&A section documents blockers formally
• compliance-gates — process-level gates that surface as blockers
• 1password — credential blockers go here first

Feedback rule inventory

Rule	Cluster	Enforcement	Last fire
feedback_blocker_surfacing	governance	manual	2026-05-01 (47-blocker audit)
feedback_question_discipline	governance	manual	2026-05-03
feedback_options_listed_vertically	governance	manual	2026-05-03
feedback_time_boxed_confirm	governance	manual	2026-05-01
feedback_scope_echo_required	governance	manual	unknown
feedback_no_partial_completion	governance	manual	2026-05-01
feedback_post_audit_persistence	governance	manual	2026-05-01
feedback_heartbeat_discipline	governance	manual	unknown

Open issues / TODOs

• No automated validator for “open blockers without SESSION-AUDIT mirror” — requires Phase 2 weekly-index-audit.timer
• 47 historical hidden blockers partially cleared but full audit not re-run; see openclaw-fragmentation-fix-2026-05-01 Phase 4.5
• HENRY-PREFERENCES.md §4 text should be cross-referenced in all plan templates; currently only in CLAUDE.md

Recent activity

• 2026-05-03: hub created (W2-S1)
• 2026-05-01: G-BLOCKER-SURFACING gate added via §A1 amendment after 47-blocker audit