forge

Abort the current forge execution and clean up. Pass a session ID (e.g. `forge abort 1`) to specify which session, or omit to auto-detect.

Brainstorm approaches for a development task before planning. Optional: pass a session ID (e.g. `forge brainstorm 2` or `forge brainstorm auth-refactor`) to use a specific session, or omit to auto-number.

Show aggregated diff of all changes since forge execution started

Execute a forge plan with parallel workers and worktrees. Pass a session ID (e.g. `forge execute 1`) to specify which session, or omit to auto-detect.

Set up forge integration with Linear and Notion for this project. Creates .forge/config.json with team, project, and page IDs.

Create a dependency-aware forge plan from requirements. Pass a session ID to use a specific session (e.g. `forge plan 1`), or omit to auto-detect.

Research codebase patterns and external docs before planning. Pass a session ID to use a specific session (e.g. `forge research 1`), or omit to auto-detect.

Roll back a forge execution to a previous checkpoint

Show current forge plan execution status and let you update task states

Validate a forge plan for integrity before execution. Pass a session ID (e.g. `forge validate 1`), or omit to auto-detect.

Run post-deploy verification on a deployed web app

Forge implementer agent — executes a single task from a forge plan. Dispatched by the executing skill via Task(). Receives a task description, file list, and working directory. Implements exactly what the task describes, nothing more.

Forge reviewer agent — reviews the implementation of a single task from a forge plan. Checks that the implementation matches the task description, only expected files were modified, and there are no obvious issues.

Cleanly abort a forge session: remove worktrees, clean up branches, mark in-progress tasks as aborted, and generate a status report.

Structured idea exploration before committing to a plan. Explores approaches, trade-offs, and constraints collaboratively. Use this BEFORE planning any non-trivial work.

Executes a forge plan batch-by-batch with parallel workers, worktrees, and review. Reads the session plan.md, dispatches implementer agents via Task(), runs reviewer agents, and tracks progress.

Reference skill for all Linear operations in forge — creating issues from tasks, updating status during execution, and importing issues as tasks. Only active when .forge/config.json exists.

Reference skill for all Notion operations in forge — pushing brainstorms and plans to Notion pages, pulling Notion content as input, and live-updating pages during execution. Only active when .forge/config.json exists.

Creates dependency-aware task plans with parallelizability analysis. Decomposes work into bite-sized tasks, builds a dependency graph, groups into execution batches, and identifies the critical path.

Optional research skill for gathering external context — library docs, best practices, patterns, and prior art. Uses WebSearch, WebFetch, Context7, and GitHub doc search. Can be invoked standalone or as part of brainstorming/planning.

Pre-execution validation of a forge plan. Checks plan integrity, file existence, git state, dependency consistency, and worktree cleanliness — without starting any agents.

Post-deploy verification for web apps. Checks URL stability (redirect loops), console errors, CORS preflight validation, and HTTP health. Use this skill AFTER deploying a web app — trigger on "verify deploy", "check deploy", "smoke test", "is the site working", or when a deploy task just completed and verification is needed. This skill is meant to be called optionally after deployments, not automatically on every task.

Reference skill for git worktree operations used during forge execution. Loaded at session start for context.

1 hook across 1 event