Sets up, resumes or maps an entire GitHub org's workspace (many repos at once): clones and onboards every repo, builds a private knowledge base of the org's systems and how they connect, connects tracker, monitoring and read-only infra, seeds a board from…
Creates, improves and measures Forge skills, including trigger accuracy, evals and benchmarks. Use when the user wants to create, edit, test or optimise a skill.
Evidence-first debugging: reproduce the failure, confirm the root cause, make the smallest fix, and prove it with a regression test. Use for bugs, crashes, stack traces and failing tests.
Creates and runs Forge project boards with the project_* tools: issues, statuses, priorities, blockers and agent runs. Use whenever the user plans, tracks or executes work as a project.
Onboards to a repository: learns how it builds, runs and tests, applies an approved minimal setup, and proves it with a smoke check. Use to set up a repo or write AGENTS.md.
Verifies a code change works: runs the tests, exercises the changed API, CLI, library or UI, and reports each claim as PASS, FAIL, BLOCKED or NOT RUN with evidence. Use to test a feature, verify a fix or check for regressions.
Evidence-verified code review of a PR, branch, worktree or uncommitted change, reporting confirmed defects separately from risks. Use to review changes before they ship.
Diagnoses and safely repairs a broken Forge setup (skills, hooks, MCP, config, providers, index). Use for setup health checks or when Forge isn't picking something up.