Skip to content

v0.9.0 — Cognitive Doctrines, Automated Upgrader & Repository Boundary Isolation

Latest
Latest

Choose a tag to compare

View all tags
@rumotion rumotion released this 21 Aug 00:38
v0.9.0
ba187da

v0.9.0 — Cognitive Doctrines, Automated Upgrader & Repository Boundary Isolation

Added

  • Automated Target Project Upgrader (scripts/upgrade-target.py): A deterministic standard-library Python tool that upgrades any target codebase with --dry-run, preserves target custom domain rules in AGENTS.md, preserves custom domain skills in .agents/skills/, updates 3-tree workflows and skills, copies safety hooks and validators, and executes post-upgrade validation automatically.
  • ARC-Skill Cognitive Doctrines Integration: Embedded battle-tested agentic cognitive patterns into AGENTS.md, docs/agent-loop.md, workflows, and skills:
    • Falsifiable Predictions: Formulate baseline failing state vs expected passing outcome before making edits ("Say what an action will do before you spend it").
    • 4-Tier Escalation Ladder: Structured diagnostic escalation (Tier 1: Surgical Inline Fix -> Tier 2: Scratch Diagnostic Script in tmp/ -> Tier 3: Diagnostic Mock Harness -> Tier 4: Model & Architecture Redesign).
    • Bimodal Execution Discipline: Single narrow probes for speculative exploration; parallel batched execution for verified plans.
    • Refuted Hypotheses Ledger (REFUTED): Explicit logging of dead theories with killing evidence to prevent repetitive retry loops across context compactions and model handoffs.
  • Repository Boundary Security & Auto-Isolation: Added strict boundary rules in AGENTS.md to make cloned templates local-only by default, preventing accidental pushes to upstream template repositories. Added scripts/detach-remote.py, auto-detach in scripts/init-fast.py, and --check-remote validator check in scripts/check-template.py.
  • Modernized Spec-Driven Development (workflows/spec-driven-development.md): Structured around executable specification anchors, characterization safety nets before legacy refactoring, and falsifiable prediction gates.
  • Updated project-upgrader Skill: Updated across all 3 discovery trees (.agents/skills/, .claude/skills/, .cline/skills/) to drive scripts/upgrade-target.py.

Changed

  • README hero image replaced with a repo-hosted, always-current infographic (SVG); docs/releasing.md now reminds maintainers to refresh front-page stats on every release.
  • Updated scripts/check-template.py with 122 required files, 9 adapters, 15 context budgets, and zero drift checks.

Fixed

  • Hero infographic caption no longer overlaps the REFLECT pill.
  • init-fast.py bootstrap prompt aligned with canonical AGENTS.md modes (dropped legacy TOKEN_SAVER wording; added atomic step-commit rule).
  • implement-task.md closes the loop: cheapest-first verify with critic pass for risky steps, plus an explicit Reflect/phase-summary step.
Assets 2
Loading