About forge
FORGE is a governed workflow skill for bounded implementation, planning, review, and delivery. It routes tasks to focused procedures, limits loaded context, and uses evidence-based checkpoints and explicit stop conditions.
Key features
- 01Task-specific workflow routing
- 02Bounded work with checkpoints
- 03Context-aware document loading
- 04Evidence and completion checks
- 05Parallel team-work guidance
Use cases
- 01Planning and implementing a scoped task
- 02Reviewing security and task evidence
- 03Preparing a release or handoff
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ redteamlife/ forge --skill forgeAfter installing, type/forgein Claude Code to confirm it loaded.
Bundled files
References and scripts installed along with SKILL.md.
assets/agent-surfaces/AGENTS.mdassets/agent-surfaces/.claude/settings.forge-fragment.jsonassets/agent-surfaces/.codex/hooks.jsonassets/agent-surfaces/.cursor/rules-scoped/go-backend.mdcjs-frontend.mdcproject-conventions.mdcpython-backend.mdcsecurity.mdcassets/agent-surfaces/.cursor/rules/forge.mdcassets/agent-surfaces/.github/copilot-instructions.mdassets/agent-surfaces/.windsurf/rules/forge.mdassets/application-docs/architecture-overview.mddeployment-playbook.mddeveloper-guide.mdincident-runbook.mdinterfaces-and-protocols.mdthreat-model.mdtool-overview.mdassets/application-docs/adr/0001-record-architecture-decisions.mdREADME.mdassets/ci-setup/github.mdgitlab.mdassets/ci/docs/commit-format.mdgovernance-patterns.mdvalidators.mdassets/ci/gitlab/security.gitlab-ci.ymlassets/ci/hooks/commit-msgpre-commitpre-pushassets/ci/policy/forge-org-policy.template.yamlDetails
From the project READMEFORGE
FORGE is an explicit workflow for bounded work: one task, clear scope, hard stops, evidence, and clean checkpoints.
Use it for governed implementation, planning, review, shipping, coordination, or context-safe bootstrap.
Routing
- Bootstrap or migrate docs:
forge-bootstrap - Plan lifecycle work:
forge-plan - Build/implement a task:
forge-buildorforge-execute-task - Review work:
forge-review - Ship, release, or close out:
forge-ship - Critique scope and assumptions:
forge-critique - Run security review:
forge-security-review - Check done/evidence:
forge-evaluation - Read/update lessons:
forge-memory - Manage private/public tool releases:
forge-tool-workflow - Coordinate multi-repo shared-contract work:
forge-cross-project
Read references only when the current task requires them:
- team and branches:
references/team-mode.md - repo flavors:
references/repo-flavors.md - agent surfaces:
references/agent-flavors.md - DevSecOps gates:
references/devsecops-gates.md - app docs:
references/application-docs.md - lifecycle map:
references/lifecycle-map.md - skill anatomy:
references/skill-anatomy.md - context and token discipline:
references/token-efficiency.md
Operating Model
- Read the minimum project-local context needed for the current step.
- Prefer
docs/forge/CONTEXT.mdwhen present; otherwise use conservativelitedefaults. - Read
docs/forge/AI.md, the compact task index or configured task source, one selected task, and task-relevant source files. - Do not load all FORGE docs, all tasks, all memory, or all security checklists at startup.
- Preserve one bounded checkpoint at a time: implementation, evidence, task state, and commit or PR/MR handoff.
- Stop on ambiguity, missing prerequisites, architecture conflict, unsafe operations, unresolved security concerns, or failed gates.
- Activation follows
activation_mode; repo-default routing requires that consent.
Team Mode
When multiple developers or agents work in parallel:
- use feature branches rather than direct work on shared branches
- claim one task before implementation
- require explicit
file_scopefor executable tasks - prefer issue assignment and labels when GitHub or GitLab is authoritative
- use append-only evidence and memory records where possible
- require CI and protected-branch merging
Output Discipline
Follow progress_policy in docs/forge/AI.md (default compact when absent):
concise checkpoint output, detailed blockers, one terminal summary — never
per-task recaps of what the task file and commit already record. Full protocol:
references/checkpoint-output.md.