Master-Claude v0.1.1

Master-Claude v0.1.1

A patch release applying fixes from automated code review (gemini-code-assist and GitHub
Copilot) on the v0.1.0 release PR. No functional changes to the tooling or knowledge base.

Fixed

• Release workflow — robust manual dispatch. release.yml now normalizes the
  workflow_dispatch tag input: a pasted refs/tags/v0.1.0 (a common GitHub UI pattern) has
  its prefix stripped, and the value is validated against the v[0-9]* shape before use. Manual
  releases no longer fail on that input. Tag-push releases are unaffected.

Changed

• Changelog accuracy. The release-workflow and README-badge entries — which shipped in the
  v0.1.0 tag — are now attributed to the [0.1.0] section instead of [Unreleased], so the
  changelog reflects what each tag actually contains.

Verification

• shellcheck --severity=warning bin/*.sh test/run.sh clean.
• sh test/run.sh passes (18 sandboxed assertions).
• Both workflow YAMLs parse; no instruction module exceeds 200 lines.

Artifacts

• master-claude-v0.1.1.tar.gz — the full installable system (tracked files at the tag).
• master-claude-v0.1.1.tar.gz.sha256 — checksum for verification.

Upgrading

No action required. v0.1.1 is drop-in; the only behavioral change is in the release workflow's
manual-dispatch path. See docs/releases/v0.1.0.md for the full feature set.

Master-Claude v0.1.0

Master-Claude v0.1.0

First public release. Master-Claude is the curated, project-agnostic synthesis of AI-agent
knowledge accumulated across ~89 codebases (Claude Code, OpenAI Codex, Google Gemini), plus a
POSIX-sh toolchain that installs that knowledge into any project and keeps it in sync — from a
single source of truth that all three agents read.

Why it exists

Agent knowledge was scattered across ~150 in-repo instruction files (97 CLAUDE.md, 13
AGENTS.md, 16 GEMINI.md, 23 CLAUDE.local.md), 165 per-project memory facts, and 63 custom
commands — duplicated, uneven, and trapped at project scope. v0.1.0 consolidates the
generalizable subset into one maintained core and makes adoption a single command.

What's included

Knowledge (docs/)

• docs/knowledge/ — the 16 recurring competency areas, clustered into 10 human-facing docs
  (commits/versioning, testing/accuracy, quality gates, docs/ADRs, architecture patterns,
  security, release ceremony, phase/sprint workflow, multi-language integration, named patterns).
• docs/architecture/ — the system's own design: distribution-model.md,
  memory-architecture.md, tri-agent-interop.md.

Distributable core (master-core/)

• AGENTS.base.md — the universal, always-applied base (the only content guaranteed to reach
  every agent in every mode).
• modules/ — 10 topic modules mirroring the knowledge areas, each curated under 200 lines.
• lang/ — concrete command overlays for rust, python, typescript, and a generic fallback.

Memory (memory-core/)

• Generalized cross-project facts in frontmatter schema v2 (adds scope, lastVerified,
  tags), with a one-directional project-to-shared promotion model and a hooks-only MEMORY.md
  index.

Assets

• commands/ — /mc-setup, /mc-setup-all, /mc-curate, plus the sprint lifecycle,
  ci-debug, bench-compare, security-audit, and mem-promote workflows.
• templates/ — a project-block scaffold and a CLAUDE.local.md session-state template.

Tooling (bin/, POSIX sh, shellcheck-clean, no heredocs)

Script	Purpose
mc-apply	Smart per-project orchestrator: detect state → install/seed/sync → verify
mc-apply-all	Workspace sweep over every project (dry-run by default)
mc-install	Build canonical AGENTS.md + CLAUDE.md/GEMINI.md symlinks; --import/--inline; seed + --trim
mc-sync	Refresh the shared core, preserving the hand-authored project block
mc-doctor	Audit: file size, symlink integrity, core drift, tri-agent parity gaps
mc-promote	Lift a project lesson into the shared memory core
mc-retrofit	Dry-run migration preview (writes nothing)
mc-commands	Register the slash commands into ~/.claude/commands

Distribution model

A project gets one canonical AGENTS.md; CLAUDE.md and GEMINI.md are symlinks to it (the
cross-tool standard — one edit reaches all three agents). Heavy modules are referenced via
@import in --import mode (Claude-primary) or baked in with --inline for full Claude/Codex/
Gemini parity. Retrofitting an existing project seeds the project block from its current
CLAUDE.md/AGENTS.md/GEMINI.md (migrated, not just backed up); --trim HTML-comments
likely-duplicate universal rules losslessly, and /mc-curate performs the semantic de-dup.

Safety properties

• Additive and reversible: existing files are backed up to *.mc-bak before replacement; all
  operations are idempotent.
• Dry-run-first for anything touching files outside the target; mc-retrofit has no apply path.
• Managed-block markers separate generated content (re-synced) from hand-authored project
  content (never overwritten).
• mc-apply refuses to run on the Master-Claude repo itself (self-hosted) without --self.

Verification

• shellcheck --severity=warning bin/*.sh test/run.sh is clean.
• test/run.sh runs 18 sandboxed assertions (install, symlinks, idempotent sync, seed + trim,
  mc-apply state detection, inline parity, retrofit safety, self-guard).
• CI (.github/workflows/ci.yml) runs lint + self-tests + a guard failing any instruction
  module over 200 lines, on every push and pull request.

Getting started

# one-time: register the slash commands
bin/mc-commands.sh --global
# then, inside any project:
/mc-setup            # detect state, install/seed, de-dup, verify
# or shell-only:
bin/mc-apply.sh /path/to/project

Artifacts

• master-claude-v0.1.0.tar.gz — the full installable system (tracked files at the tag).
• master-claude-v0.1.0.tar.gz.sha256 — checksum for verification.

GitHub also attaches auto-generated Source code archives.

Compatibility & requirements

• POSIX sh; tools used: git, sed, awk, grep, ln, mktemp (standard on Linux/macOS).
• shellcheck for linting; gh CLI for the release workflow. No language runtime required.

Known limitations

• @import is Claude-specific; Codex/Gemini get the inline base in --import mode (use
  --inline for full parity).
• mc-retrofit is dry-run only in this release; applying a retrofit uses mc-install/mc-apply.
• Semantic de-dup (/mc-curate) requires an LLM session; the deterministic --trim is the
  shell-only approximation.

Links

• Overview: docs/00-OVERVIEW.md
• Taxonomy: docs/01-KNOWLEDGE-TAXONOMY.md
• Distribution model: docs/architecture/distribution-model.md
• Changelog: CHANGELOG.md
• Contributing: CONTRIBUTING.md