Forgeplan Workflow Plugin

What It Does

The forgeplan-workflow plugin adds the /forge command to Claude Code and compatible AI agents. It runs the full Forgeplan methodology cycle in a single conversational command:

/forge "Add payment processing"

This triggers: Route -> Shape -> Validate -> Code -> Evidence -> Activate.

Installation

Via marketplace (npx)

npx skills add ForgePlan/marketplace --skill forge

Via built-in CLI (offline, no network)

If you already have the forgeplan binary installed:

forgeplan setup-skill

This writes the embedded skill file to ~/.claude/skills/forge/SKILL.md. No network access, no marketplace fetching — the skill definition is compiled into the binary. See forgeplan setup-skill.

Commands

Command	Description
`/forge "task"`	Full cycle: route -> create -> validate -> build -> evidence -> activate
`/forge-cycle`	Explicit step-by-step forge cycle with 8 phases
`/forge-audit`	Multi-expert code audit with methodology integration

How /forge Works

When you invoke /forge "Add rate limiting", the skill:

Route — calls forgeplan_route to determine depth (Tactical / Standard / Deep / Critical)
Shape — creates the right artifact (PRD, RFC, etc.) via forgeplan_new
Validate — checks quality gates via forgeplan_validate
Reason — runs ADI reasoning if Standard+ depth (3+ hypotheses)
Code — builds the solution with tests
Evidence — creates evidence pack, links to artifact, checks R_eff
Activate — marks artifact as active via forgeplan_activate

For Tactical depth, the skill skips artifacts and just executes the task directly.

/forge-cycle — Explicit Step-by-Step

When you need more control over each phase:

/forge-cycle PRD-001

Runs 8 explicit phases:

Phase	Action
0. OBSERVE	`forgeplan health` — understand project state
1. ROUTE	Determine depth and pipeline
2. SPRINT	Plan implementation waves
3. BUILD	Implement the solution
4. AUDIT	Adversarial multi-expert review
5. FIXES	Fix HIGH/CRITICAL findings
6. EVIDENCE	Create evidence, link, score
7. COMMIT	Git commit + PR

/forge-audit — Methodology-Aware Audit

Combines code auditing with Forgeplan’s quality framework. Reports findings against both code quality and methodology compliance (missing evidence, unlinked artifacts, R_eff gaps).

Agentic RAG Architecture

The skill uses agentic RAG — it loads only the relevant portion of the methodology knowledge base for each step (~300 lines), not the entire specification. The SKILL.md file acts as a router:

Maps user needs to specific methodology sections
Provides MCP tool reference (which forgeplan_* tool to call)
Includes depth calibration rules and escalation triggers
Documents the full artifact lifecycle

Knowledge base sections included

Section	Content
MCP tool table	All `forgeplan_*` tools with CLI equivalents
Core workflow	6-step cycle: health -> route -> new -> validate -> review -> activate
Depth calibration	Tactical / Standard / Deep / Critical decision matrix
Evidence rules	Structured fields, CL scoring, R_eff calculation
Proactive behavior	When to escalate, when to suggest artifacts
Lifecycle states	Draft -> active -> superseded/deprecated flow

Worked Example

User: /forge "add rate limiting to the API"

Agent: [calls forgeplan_route("add rate limiting to the API")]
  -> Depth: Standard, Pipeline: PRD -> RFC

Agent: [calls forgeplan_new(kind: "prd", title: "API Rate Limiting")]
  -> Created PRD-042

Agent: [fills Problem, Goals, Non-Goals, FR sections]

Agent: [calls forgeplan_validate("PRD-042")]
  -> PASS (0 MUST errors)

Agent: [implements rate limiting middleware + tests]

Agent: [calls forgeplan_new(kind: "evidence", title: "Rate limit tests -- 8 pass")]
  -> Created EVID-089

Agent: [calls forgeplan_link("EVID-089", "PRD-042", "informs")]
Agent: [calls forgeplan_activate("PRD-042")]
  -> draft -> active, R_eff = 1.00