By chkdsklabs
Agentic design workflow: April elicits the five upstream design documents, Shredder slices them into a dependency-ordered queue, and Karai dispatches Bebop / Baxter / Chaplin / Metalhead / Splinter / Tatsu / Krang through Bishop review and Tiger Claw adversarial QA, with Traag-style scope enforcement via PreToolUse hook.
Run the harness-owned mutagen execution loop until the queue clears, stalls, or escalates.
Invoke Traag to evaluate an in-flight scope-manifest amendment request. Traag decides ALLOW or DENY against the active slice's stage, agent, and the global denylist.
Turn the Bishop advisory backlog into a cleanup slice (or series) by dispatching Shredder against .mutagen/state/advisory-backlog.jsonl.
Run April to interview the user and author the five upstream design documents (PRD, ADR, DDD, ISC, DSD).
Pause / resume / status the harness execute-next loop at the next stage boundary.
As the reporter 'April O'Neil', you interview the user and author the five upstream design documents (PRD, ADR, DDD, ISC, DSD) from the templates in this repo. Mark unknowns as `<TBD>`; never invent domain details; watch for cross-document contradictions. You don't write code, tests, infrastructure, or reviews.
As 'Baxter', you execute math-heavy, algorithmic, or deep-reasoning slices — recursion, combinatorics, numerical analysis, spatial geometry, optimization, formal correctness. Produce mathematically sound code that upholds every cited invariant. Standard plumbing goes to Bebop; infrastructure goes to Krang.
As 'Bebop', you execute standard slices — CRUD, UI, business logic, middleware, migrations, general Layer 2–5 plumbing. You're the muscle: write the code, uphold cited invariants, update state. You don't question the architecture.
As 'Agent Bishop', you review completed slice artifacts at the principal-engineer level: design smells, abstraction quality, cognitive load, naming, error/logging discipline, API ergonomics, performance smells, change hygiene, pattern consistency. Catalogue findings by severity and return a verdict. You never rewrite code; you may attach Suggested Fixes for the author's retry. You sit outside the Foot Clan by design.
As 'Dr. Chaplin', you own non-trivial Layer 2 (Data) slices and data-migration Layer 6 slices. Data Model Analysis before any schema change; indexes and partitioning justified by query; online migrations with backfill and rollback. Application logic goes to Bebop; database provisioning goes to Krang.
Explicit-only skill. Resume a slice after a manual repair without manually re-running structural-check, update-queue, transition-active-slice, and dispatch-stage in sequence. Use when an in-flight slice was hand-repaired (an author dispatch wrote a partial artifact, the operator fixed the file on disk, and the queue should now resume as if the repair was the canonical author output). Invoke only when the user explicitly says $mutagen-resume.
Explicit-only skill. First-run Pushover configuration wizard — detect current state, collect user key and app token, choose storage (env vars or workflow.json), optionally configure quiet events, send a test notification. Invoke only when the user explicitly says $mutagen-setup-pushover.
Explicit-only skill. Run Shredder against the approved upstream design bundle (PRD / ADR / DDD / ISC / DSD) to produce a dependency-ordered slice queue. Invoke only when the user explicitly says $mutagen-slice. Do not trigger on ambient mentions of "slice" or "slicing".
Explicit-only skill. Read-only report on the mutagen workflow — upstream document status, April's Readiness Brief, Shredder's Validation Report, pipeline mode, slice queue progress, harness queue validation, active slice, heartbeat telemetry, gate telemetry, open escalations, recent reviews. Invoke only when the user explicitly says $mutagen-status.
Explicit-only skill. Invoke Traag to evaluate an in-flight scope-manifest amendment request. Traag decides ALLOW or DENY against the active slice's stage, agent, and the global denylist. Invoke only when the user explicitly says $mutagen-amend-scope and provides the path (or glob) to add, mutation kind, and a reason.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
A Rust harness that runs an agentic design workflow on top of Claude Code and Codex CLI. Mutagen owns the things prompts can't be trusted with — queue selection, stage transitions, scope enforcement, evidence bundling, retry policy, and persisted verdicts — and hands the rest to a fixed cast of personas (April → Shredder → Karai → Bebop / Baxter / Krang / Chaplin / Tatsu / Metalhead → Bishop → Tiger Claw → Splinter). The pipeline turns a five-document upstream design bundle (PRD / ADR / DDD / ISC / DSD) into dependency-ordered slices, dispatches each slice to the right executor, and gates the result on adversarial review before it touches the queue.
If a behavior matters, the harness enforces it or records it. If the only control is "the prompt said pretty please," that is not a control plane.
Mutagen ships as a plugin so the same workflow drops into both supported hosts.
/plugin marketplace add CHKDSKLabs/Mutagen
/plugin install mutagen@mutagen-marketplace
Verify with /plugin marketplace list and /plugin list.
Clone this repo, then register the marketplace entry at
~/.agents/plugins/marketplace.json (or use the repo-local
.agents/plugins/marketplace.json that ships here), pointing at
plugins/mutagen/. Export MUTAGEN_ROOT so skill bodies can resolve it:
export MUTAGEN_ROOT="/absolute/path/to/Mutagen/plugins/mutagen"
Codex auto-discovers skills under plugins/mutagen/skills/<name>/SKILL.md.
Invoke with $mutagen-slice, $mutagen-execute-next, $mutagen-status,
$mutagen-amend-scope, $mutagen-elicit, $mutagen-consolidate-advisories,
$mutagen-setup-pushover, $mutagen-pause, or $mutagen-resume. All nine
skills are configured with allow_implicit_invocation: false — Mutagen is a
workflow, not a helpful tool, so explicit invocation is the only trigger.
Known degradation on Codex: the codex_hooks feature is still under
development and disabled on Windows, so the plugin doesn't ship manifest-level
hooks. Scope manifests are written between stages for audit and visibility, but
enforcement is advisory (the agent is told its allowed globs; nothing blocks
it). Reviewers are the backstop.
PreToolUse scope-enforcement hook (Claude Code only) that blocks
writes outside the active slice's manifest before they happen.mutagen-harness) that owns queue mutation,
evidence assembly, structural checks, retry policy, and verdict persistence —
so behavior doesn't change just because you switched hosts.For the full feature surface see plugins/mutagen/README.md.
For harness internals see harness/README.md.
For a populated reference workspace — five upstream design documents, a slice
queue, and a Tiger Claw review report in their canonical filesystem layout —
see examples/orders-demo/.
.
├── .claude-plugin/marketplace.json # Claude Code marketplace manifest
├── .agents/plugins/marketplace.json # Codex marketplace registration
├── plugins/mutagen/ # the plugin (dual-host)
│ ├── .claude-plugin/plugin.json
│ ├── .codex-plugin/plugin.json
│ ├── agents/ # 13 personas
│ ├── commands/ # 9 Claude slash commands
│ ├── skills/ # 9 Codex skills ($mutagen-*)
│ ├── bin/ # host adapters + harness launcher
│ ├── hooks/ # Claude PreToolUse + PostToolUse
│ ├── scripts/ # wrappers, dispatch glue, notify
│ ├── templates/ # PRD / ADR / DDD / ISC / DSD
│ └── guides/ # authoring & review guides
├── harness/ # mutagen-harness Rust crate
└── examples/orders-demo/ # populated reference workspace
claude plugin validate .
Codex has no equivalent validator today; manually confirm
plugins/mutagen/.codex-plugin/plugin.json parses and that every skill
directory contains a SKILL.md.
MIT.
npx claudepluginhub chkdsklabs/mutagen --plugin mutagenP2P sync for your .claude configuration across machines. Like OneDrive, but for crabs.
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.
Comprehensive PR review agents specializing in comments, tests, error handling, type design, code quality, and code simplification
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.