plugin-forge

Plugin plugin-forge from marketplace claude-dev-toolkit in Aznatkoiny/claude-dev-toolkit. Commands read from the repo's manifest — 1 month ago.

Skills (11)

• arm-evals

  Arm a forge eval suite into a frozen build contract. Run when .forge/state.json shows phase=evals, after /plugin-forge:build-evals; re-run after any suite edit or triage grader-error to re-version and re-freeze. Four gates in order via forge-eval: reference (every task's reference solution passes it

• build-evals

  Generate a runnable eval suite (tasks, graders, judges, targets) for a Claude Code plugin BEFORE it is built. Use for "build evals", "eval suite for my plugin", "test my plugin", "retrofit evals", "mine traces into tasks". In the forge pipeline this is the evals phase; retrofit and mine modes run st

• build-loop

  Launch the isolated build-until-green loop for a forge project. Run when .forge/state.json shows phase=armed, after /plugin-forge:arm-evals. Walks the preconditions checklist (trust dialog, hooks enabled, MCP OAuth preflight for live servers, worktree base ref, permissions.allow for a long unattende

• forge

  Build a new Claude Code plugin eval-first — interview, design contract, frozen eval suite, headless goal-loop build, verified ship. Use for "build me a plugin", "plugin for X", "forge a plugin", "resume forge", "forge status". Single entry point for the plugin-forge pipeline; resumable across sessio

• generation-standards

  House style for every file plugin-forge emits into a generated plugin. Load before writing or reviewing any generated SKILL.md, agent, hooks.json, hook script, .mcp.json, plugin manifest, marketplace entry, README, or settings snippet — and before ship runs the freshness-guard sweep. Covers descript

• primitive-catalog

  Decision cards for selecting Claude Code primitives while designing a plugin - skill vs hook vs subagent vs agent team vs MCP server vs bundled script vs LSP vs workflow vs loop. Load during any plugin architecture or composition discussion, when writing a Primitive Decision Record (design/CONTRACT.

• prompt-optimizer

  Optimize, rewrite, or create prompts for Claude models. Use when the user wants to improve an existing prompt, turn a rough thought dump into a well-structured prompt, write a prompt from scratch, craft a Deep Research request, or tune a prompt for a specific model (Fable 5, Mythos 5, Opus 5, Opus 4

• ship

  Package and release a verified forge-built plugin: claude plugin validate --strict, frontmatter/scoped-name/freshness lints, README with the permission story, versioning decision (SHA vs pinned semver), distribution menu (skills-dir, private marketplace, community), and installing the trace-capture

• skill-forge

  Research-driven skill creation workflow. Guides you through intent capture, web research, prompt optimization, multi-agent deep research, and skill packaging — all grounded in real sources, not training data. Also invoked by plugin-forge's write-contract and build-evals for domain grounding.

• verify

  Verify a forge-built plugin before shipping: smoke matrix, fresh pass^k trials, holdout tranche, trigger evals on default AND crowded targets, token budget check, and triage of sampled failures. Use after a forge build loop has gone green (.forge/state.json phase smoke or verify) and before /plugin-

• write-contract

  Author the Primitive Decision Record (design/CONTRACT.md) during the forge pipeline's contract phase: per-capability primitive compositions with six-step consequence chains (trigger, context economics, enforcement, failure & degradation, testability, distribution), at least one rejected alternative

Subagents (2)

• spec-critic

  Use proactively after a plugin contract (design/CONTRACT.md) or eval suite draft exists and before any build begins. Audits the draft for ambiguity, grader defects, unobservable primitives, and reward-hack surfaces through three lenses (context economics, headless/enforcement, distribution/maintenan

• triage

  Use proactively after an eval run completes with failures. Samples failing (and suspicious passing) trials from runs/<ts>/ and classifies each failure as agent-error, grader-error, or task-ambiguity using dual-trajectory evidence — the agent transcript read side by side with the grader's evidence an

Hooks (4)

• PreToolUse

  ${CLAUDE_PLUGIN_ROOT}/hooks/scripts/guard-evals.sh

• PreToolUse

  ${CLAUDE_PLUGIN_ROOT}/hooks/scripts/guard-evals.sh

• Stop

  ${CLAUDE_PLUGIN_ROOT}/hooks/scripts/sweep-evals.sh

• SessionStart

  ${CLAUDE_PLUGIN_ROOT}/hooks/scripts/reinject-context.sh

Similar plugins