Skip to content

v1.5.0 — Engine-layer governance + orchestration codification

Choose a tag to compare

View all tags
@MrBinnacle MrBinnacle released this 27 May 21:55
· 11 commits to master since this release
v1.5.0
6b4d7ff

Engine-layer governance (canonical spec + always-loaded runtime mirror)

  • BEHAVIOR_SPEC.md authored as the canonical engine specification. §1–§8 reduce all engine decision authority — intake routing, mode selection, reference loading, verdict conditions, module firing, confidence bounds, behavioral overrides, and domain-policy gating — into deterministic IF/THEN rules. [AMBIGUOUS-EXTRACTED] tags carry forward pre-existing ambiguities from source material unresolved.
  • SKILL.md ↔ BEHAVIOR_SPEC.md mirror invariant established. SKILL.md is the always-loaded runtime mirror; BEHAVIOR_SPEC.md is the canonical spec. Every load-bearing rule MUST exist in both files with SKILL.md carrying the full text inline (the 5 load-bearing behavioral rules, intake routing consequents, mode selection triggers, verdict conditions, reference-loading matrix). This preserves v1.4.0's partial-load discipline (0/5 → 5/5 rule survival) while letting maintainers reason from a single canonical source. The mirror invariant is declared in both files' preambles and codified in docs/MAINTENANCE.md.
  • Decision authority stripped from diagnostics/ and references/. Each file's preamble now points decision authority to BEHAVIOR_SPEC.md; verdict consequents removed; observational catalogs preserved. references/mode-behaviors.md and references/module-guide.md reduced (~37% size) — rule machinery moved to spec, module bodies kept as rationale.
  • docs/VALIDATION.md authored. Enforcement spec for the four repo-integrity rules (description char-count, gotchas section-count, banned-token, SKILL.md path integrity). The PreToolUse git commit hook mirrors this spec; if they diverge, the spec is authoritative.

Namespace topology + presentation/policy separation

  • templates/ renamed to domain-policies/. Resolves ADR-0002 Issue C (the directory's actual contract is domain gating/configuration, not formatting scaffolding). All 8 domain policies retain their -azimuth.md filenames.
  • Domain-policies cite spec rules per RULE-8.10. Every rule-shaped sentence in domain-policies/*-azimuth.md now explicitly cites the BEHAVIOR_SPEC.md rule it activates, suppresses, or parameterizes. Three domain-level confidence-cap statements replaced with citations to the cross-cutting RULE-6.5 (incentive-misalignment cap) per RULE-8.6.
  • .claude/skills/ namespace split (narrowed). Four maintenance skills (research-scout, verdict-auditor, gap-scanner, reference-authoring) moved under .claude/skills/maintenance/. Speculative runtime/, eval/, plugin/, meta/ namespaces deliberately NOT created — empty namespaces would imply ontology without occupants.
  • ui-ux-pro-max plugin skill removed as superseded by the global impeccable skill. Local install had been half-broken (data/scripts pointed at a non-existent src/ tree). .claude/skills/ now contains only maintenance/ plus the gitignored azimuth/ install-test self-copy.
  • executive-azimuth.md relocated. Identified as a presentation/format file misfiled under domain-policies/. Moved to references/output-format-executive.md. Domain-policies/ now contains only true domain-gating policies (8 files, all wired through BEHAVIOR_SPEC.md §1.3 RULE-1.12..RULE-1.19).

Spec corrections

  • RULE-6.6 narrowed. Layer-3 domain set reduced from {3 (Hiring), 8 (Startup)} to {8 (Startup)} only — the co-founder governance structural-completeness rule has no source support in domain-policies/hiring-azimuth.md. Spec previously claimed cross-domain firing; verification of source files showed only startup-azimuth.md defines a Co-Founder Structure Check.
  • RULE-8.12 stripped of source-unsupported clause. Removed the "PROCEED suppressed per §6 RULE-6.6 when co-founder governance structure incomplete (applies to co-founder hires only)" clause. This was an inferred extension to the Hiring domain with no source support.
  • RULE-4.15 'fragility-CRITICAL' label dropped. Eliminates a cross-layer reference leak (the label appeared only in diagnostics/fragility-scan.md, not in spec).

Maintenance orchestration

  • ADR-0001 added. docs/adr/0001-bespoke-orchestration-layer.md records the producer/consumer split, lazy file creation, hard-dep vs soft-dep maintenance skills, and what is deliberately not built (no pipeline skill, no verify-mode flag, no upfront CONTEXT.md, no ADR backfill).
  • ADR-0002 added. docs/adr/0002-engine-layer-governance.md declares the SPEC / RUNTIME / DIAGNOSTICS / RATIONALE four-layer taxonomy. Both ADRs subsequently aligned with executed state (post-execution footers added).
  • docs/MAINTENANCE.md authored. Operational orchestration anchor: verification layers (commit-time / audit-time / release-time), maintenance-loop responsibilities, subagent dispatch invariants, migration / rollback discipline (including the path-rename follow-up-commit pattern), drift-vectors table, and 6-step next-session continuation. Reading it + the two ADRs is the documented onboarding path for any external maintainer.
  • .out-of-scope/ directory seeded. Two rejection-rationale files promoted from private maintainer notes to durable, citable repo artifacts: template-expansion-without-evidence.md and breadth-before-reliability.md.
  • docs/agents/domain.md updated. Now points consumer skills at docs/adr/ as the durable record; CHANGELOG remains the record for shipped behavioural changes.
  • reference-authoring skill added (fourth maintenance skill) with discipline for new reference files and domain-policies: EXTEND-vs-CREATE, Module 7 vocabulary header, sourcing caveat, pre-verdict gate.

Repo hygiene + fork legibility

  • .gitignore adjustments.
    • docs/adr/ and docs/agents/ ship with the repo so forks get a working orchestration substrate.
    • docs/MAINTENANCE.md added to the docs allow-list.
    • .claude/skills/azimuth/ (local install-test self-copy) and .playwright-mcp/ (MCP cache) ignored.
    • CLAUDE.md removed from .gitignore (the file was tracked but the entry implied "internal only" — implementation contradicted intent; resolved by treating CLAUDE.md as the public maintainer guide it already was).
  • BEHAVIOR_SPEC.md fork-legibility fix. Dropped the dangling session-local rationale reference (docs/superpowers/specs/…) that pointed at a gitignored artifact. ADR-0002 is now the sole rationale anchor.
  • README, CLAUDE.md, gap-scanner, ADRs aligned with the templates/→domain-policies/ rename, the .claude/skills/maintenance/ namespace split, and the executive-format relocation. No dangling paths remain in tracked .md content (CHANGELOG historical entries and evals/test-*-control-skill.md frozen snapshots intentionally preserved).
  • evals/README.md annotated. Explains why test-*-control-skill.md files reference pre-rename templates/... paths (frozen control snapshots; modifying invalidates baselines) and names the correct path forward (capture a new version-tagged file).
  • Two role-bound subagents added. .claude/agents/azimuth-evidence-checker.md audits proposed numeric heuristics / thresholds / empirical claims; .claude/agents/azimuth-output-auditor.md stress-tests pasted AZIMUTH outputs against the skill's structural rules. Both read-only; both invariants documented in docs/MAINTENANCE.md.

Planned

See ROADMAP.md for the why on each.

  • Evidence tags. Every output claim carries a provenance tag.
  • Audit trail. Output lists modules run and what each surfaced.
  • Market-entry template. Dedicated template for geographic expansion / international market-entry decisions (post-build eval identified product-launch-azimuth.md as an imprecise proxy for this domain).
Assets 2
Loading