RooForge

A curated set of AI agent skills for orchestrating large efforts on an issue tracker — from a loose idea to a merged PR. The repo defines the skills; the forge drives the flow; the tracker carries the map.

Overview

The repo has two layers:

skills/ — 35 vendor-agnostic skills an agent loads mid-session. Communication skills (caveman, ste100, conventional-commits) govern all text. Orchestrator skills (forge, forge-flow, loops) drive the session. Planning skills (wayfinder, grilling, prototype, deep-research, planning-and-task-breakdown, domain-modeling) shape the map. Execution skills (using-git-worktrees, subagent-driven-development, dispatching-parallel-agents, finishing-a-development-branch, verification-before-completion, pr-review, pr-resolve, creating-pull-requests) build the code. Bootstrap skills (forge-init, forge-setup, forge-docs, forge-cleanup) maintain the repo itself. Sidecar skills (12-factor-app, kiss-principle, frontend-design, forge-tailwindcss-conventions, forge-eu-accessibility, forge-seo) ride alongside — domain conventions loaded when the task touches them.
docs/ — convention scaffolding for the artefacts the agent produces: ADRs in docs/adr/, system designs in docs/system-design/, how-to guides in docs/guides/, deep research artifacts in docs/dev/agents/. RFC-style decisions live as wayfinder tickets on the tracker, not as docs files.

Token budget

Counts use OpenAI o200k_base, which Tiktokenizer marks compatible with GPT-5.6.

Scope	Count
Session-start `SKILL.md` bodies (`forge`, `forge-flow`, `caveman`, `wayfinder`)	7,547 tokens
Session-start `name` and `description` values	262 tokens
All 35 `SKILL.md` bodies, excluding frontmatter	39,018 tokens
All 35 `name` and `description` values	1,188 words, 1,803 tokens

Counts cover prompt text only. They exclude tool schemas, chat framing, and harness overhead.

Install

Copy this prompt into a new agent session:

Download the forge-setup skill from https://raw.githubusercontent.com/weselben/RooForge/main/skills/forge-setup/SKILL.md.

Install it at `<harness-config-dir>/skills/forge-setup/SKILL.md`. If no harness-specific skill directory exists, install it at `~/.agents/skills/forge-setup/SKILL.md`.

Load the skill and follow its procedure exactly. Use a temporary clone for discovery and patches. Ask before changing my harness configuration.

After forge-setup completes, invoke forge-init by name once per repo. forge-init is user-invoked. Then start working with a prompt like "work on the next ticket" or "let's continue the auth refactor". The always-loaded forge skill invokes forge-flow and runs the full flow.

Prerequisites

The skills in this repo have the following external dependencies. Ensure these are available on your system before running the corresponding skills.

Skill	Required tools / dependencies
forge, forge-flow, wayfinder, git-issue-tracker, creating-pull-requests, pr-review, pr-resolve, finishing-a-development-branch	`gh` (GitHub CLI, ≥2.0), authenticated (`gh auth status`)
forge-init, wayfinder	`gh` + `gh api` (label creation, sub-issue wiring)
loops, pr-review, pr-resolve	`bash` (POSIX), `kimi -p` (or harness equivalent after forge-setup)
using-git-worktrees, subagent-driven-development, finishing-a-development-branch	`git` (≥2.20 for `git worktree`), project test runner (`npm test` / `cargo test` / `pytest` / `go test ./...`)
using-git-worktrees	Project package manager if `package.json` exists (`npm`/`pnpm`/`yarn`)
deep-research	Web search capability (harness tool or `curl` + `jq` for API fallback)
forge-eu-accessibility, forge-seo	Web search + URL fetch capability (e.g. `WebSearch` / `FetchURL`) for live legal / standards verification
dispatching-parallel-agents, subagent-driven-development	Harness subagent API (`AgentSwarm`, `Agent`, `run_in_background`, `subagent_type: coder/explore/plan`) — adapted by forge-setup
forge-flow	Harness goal API (`CreateGoal`) or file fallback (`GOAL.md`) — adapted by forge-setup
planning-and-task-breakdown, forge	Harness plan mode (`EnterPlanMode`/`ExitPlanMode`) — adapted by forge-setup
caveman, caveman-commit, caveman-review, ste100, conventional-commits, grilling, prototype, domain-modeling, verification-before-completion, resolving-merge-conflicts, use-git-identity, forge-init, forge-docs, forge-cleanup, forge-setup, 12-factor-app, kiss-principle, frontend-design, forge-tailwindcss-conventions	No external tools (pure prompt contracts)

Note: Skills marked adapted by forge-setup reference Kimi-specific machinery (CreateGoal, EnterPlanMode, AgentSwarm, kimi -p). The forge-setup skill (run once after cloning on a non-Kimi harness) discovers and patches these to the harness's equivalents. Until forge-setup runs, those skills will reference unavailable Kimi APIs.

The Flow

Forge owns a single path: map → resolve → plan → work → verify → review → resolve. Each step ends on a checkable criterion; the next step is the proof. forge is the always-loaded orchestrator — it auto-triggers on session prompts like "work on the next ticket" and invokes forge-flow once at session start to bootstrap branch and goal before step 1.

Each step is driven by one skill and supported by others:

Step	Driving skill	Supporting skills
0. Bootstrap	`forge-flow`	`ste100`
1. Map	`wayfinder`	`git-issue-tracker`, `grilling`, `domain-modeling`, `deep-research`, `dispatching-parallel-agents`
2. Resolve	the ticket's type skill (`grilling` / `prototype` / `deep-research` / task)	`domain-modeling` (sweep after every close)
3. Plan	`planning-and-task-breakdown`	—
4. Work	`subagent-driven-development`	`using-git-worktrees`, `dispatching-parallel-agents`, `conventional-commits`, `caveman-commit`, `forge-docs`, `resolving-merge-conflicts`
5. PR	`creating-pull-requests`	`ste100`
6. Verify	`verification-before-completion`	—
7. Review	`pr-review`	`loops`, `caveman-review`
8. Resolve findings	`pr-resolve`	`use-git-identity`, `loops`, `resolving-merge-conflicts`

Always-on regardless of step: forge (orchestrator), caveman (ultra), wayfinder.

1. Session Start — Forge loads, forge-flow bootstraps

The user opens a chat and writes something like "work on the next ticket", "let's continue the auth refactor", or simply describes a task. The forge skill is always-loaded and auto-triggers on these prompts. It invokes forge-flow once at session start as a frontier sub-skill: forge-flow detects whether a wayfinder map exists, prepares the feat branch from main, writes the long-living contract goal, and hands off to forge step 1. Then forge's own step 1 either loads the existing map or charts a new one through wayfinder.

flowchart TD
    subgraph Forge["<b>forge</b> — always-loaded orchestrator"]
        direction TB
        F1["Step 1: Map<br/><b>wayfinder</b>"]
        F2["Step 2: Resolve<br/>ticket-type skill"]
        F3["Step 3: Plan<br/><b>planning-and-task-breakdown</b>"]
        F4["Step 4: Work<br/><b>subagent-driven-development</b>"]
        F5["Step 5: PR<br/><b>creating-pull-requests</b>"]
        F6["Step 6: Verify<br/><b>verification-before-completion</b>"]
        F7["Step 7: Review<br/><b>pr-review</b>"]
        F8["Step 8: Resolve findings<br/><b>pr-resolve</b>"]
        F1 --> F2 --> F3 --> F4 --> F5 --> F6 --> F7 --> F8
    end

    Start([Session Start<br/>user prompt]) --> Forge
    Start --> FF["<b>forge-flow</b><br/>session bootstrap (frontier sub-skill)"]
    FF -->|"detect map · name branch<br/>write goal · hand off"| F1

    style Start fill:#95A5A6,color:#fff,stroke:#7F8C8D
    style Forge fill:#16A085,color:#fff,stroke:#0E6655
    style FF fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style F1 fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style F2 fill:#F39C12,color:#fff,stroke:#B8750E
    style F3 fill:#F39C12,color:#fff,stroke:#B8750E
    style F4 fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style F5 fill:#2C3E50,color:#fff,stroke:#1A252F
    style F6 fill:#27AE60,color:#fff,stroke:#1A7A42
    style F7 fill:#E67E22,color:#fff,stroke:#A05A15
    style F8 fill:#E67E22,color:#fff,stroke:#A05A15

forge-flow does not load the map itself — it only detects whether one exists. The actual map load happens in forge step 1 (section 2 below).

2. Map — Load or Chart the Wayfinder Map

The map is a single issue on the tracker, labelled wayfinder:map. Its tickets are child issues; each ticket carries a wayfinder:<type> label naming the skill that resolves it. If a map already exists, wayfinder loads it; if not, it enters chart mode (grilling + domain-modeling) and creates one. Sub-skills involved: git-issue-tracker for the GitHub operations, grilling to draw out the destination, domain-modeling to seed the glossary, deep-research for AFK research tickets, and dispatching-parallel-agents to chart fog patches in parallel.

flowchart LR
    subgraph Map["Wayfinder Map"]
        direction TB
        MapIssue["Map Issue<br/>wayfinder:map"]
        Tickets["Decision Tickets<br/>wayfinder:research | prototype<br/>grilling | task | domain-modeling"]
    end
    
    subgraph Skills["Resolution Skills"]
        direction TB
        Research["deep-research<br/>(AFK, parallel)"]
        Prototype["prototype<br/>(HITL)"]
        Grilling["grilling<br/>(HITL)"]
        Task["task<br/>(HITL or AFK)"]
        DomainModel["domain-modeling<br/>(glossary + ADRs)"]
    end
    
    MapIssue -->|"child issues"| Tickets
    Tickets -->|"wayfinder:research"| Research
    Tickets -->|"wayfinder:prototype"| Prototype
    Tickets -->|"wayfinder:grilling"| Grilling
    Tickets -->|"wayfinder:task"| Task
    Tickets -->|"wayfinder:domain-modeling"| DomainModel
    
    Grilling -->|"after close"| DomainModel
    DomainModel -->|"glossary"| Context["docs/dev/CONTEXT.md"]
    DomainModel -->|"ADR"| ADR["docs/adr/NNNN-*.md"]
    
    style MapIssue fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style Tickets fill:#F39C12,color:#fff,stroke:#B8750E
    style Research fill:#27AE60,color:#fff,stroke:#1A7A42
    style Prototype fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style Grilling fill:#E67E22,color:#fff,stroke:#A05A15
    style Task fill:#16A085,color:#fff,stroke:#0E6655
    style DomainModel fill:#E67E22,color:#fff,stroke:#A05A15
    style Context fill:#2C3E50,color:#fff,stroke:#1A252F
    style ADR fill:#2C3E50,color:#fff,stroke:#1A252F

3. Resolve — Work One Ticket

One ticket per session. wayfinder claims the next frontier ticket (assigns it), the ticket's wayfinder:<type> label names the resolution skill (grilling, prototype, deep-research, or a task), and on close domain-modeling sweeps for new glossary terms and decisions worth an ADR. Sub-skills: dispatching-parallel-agents may run multiple research tickets in parallel; the next session picks up the next frontier ticket or, when the frontier empties, hands off to step 3 (plan).

flowchart LR
    Ticket["Open Ticket<br/>wayfinder:<type>"] --> Claim["<b>wayfinder</b><br/>claim (assign)"]
    Claim --> Skill["<b>grilling</b> / <b>prototype</b> /<br/><b>deep-research</b> / task<br/>per type label"]
    Skill --> Resolve["Resolve<br/>post answer, close issue"]
    Resolve --> Sweep["<b>domain-modeling</b><br/>glossary + ADR sweep"]
    Sweep --> Pointer["<b>wayfinder</b><br/>append pointer to<br/>map Decisions-so-far"]
    Pointer --> Fog{"New fog<br/>specifiable?"}
    Fog -->|"yes"| NewTickets["<b>wayfinder</b><br/>create new tickets<br/>graduate from Not yet specified"]
    Fog -->|"no"| Frontier{"Frontier<br/>empty?"}
    NewTickets --> Frontier
    Frontier -->|"no"| NextTicket["Next session:<br/>next ticket"]
    Frontier -->|"yes"| Plan["Step 3: <b>planning-and-task-breakdown</b>"]
    
    style Ticket fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style Claim fill:#F39C12,color:#fff,stroke:#B8750E
    style Skill fill:#E67E22,color:#fff,stroke:#A05A15
    style Resolve fill:#27AE60,color:#fff,stroke:#1A7A42
    style Sweep fill:#E67E22,color:#fff,stroke:#A05A15
    style Pointer fill:#16A085,color:#fff,stroke:#0E6655
    style Fog fill:#F39C12,color:#fff,stroke:#B8750E
    style NewTickets fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style Frontier fill:#F39C12,color:#fff,stroke:#B8750E
    style NextTicket fill:#95A5A6,color:#fff,stroke:#7F8C8D
    style Plan fill:#2C3E50,color:#fff,stroke:#1A252F

4. Work — Subagent-Driven Development

When the frontier empties, planning-and-task-breakdown breaks the cleared map into ordered tasks and the user approves the plan in plan mode. Then subagent-driven-development takes over as coordinator: it creates one worktree per task via using-git-worktrees (under .worktrees/<task-slug>/), dispatches all implementer subagents in a single dispatching-parallel-agents swarm call (each runs using-git-worktrees + conventional-commits + caveman-commit + verification-before-completion), reviews each task's diff, runs the fix loop until clean, and squashes each task branch into one Conventional Commit on the feat branch (git merge --squash then git commit). Conflicts route through resolving-merge-conflicts. After every squash, forge-docs updates the docs indexes.

flowchart LR
    Plan["Plan Approved<br/><b>planning-and-task-breakdown</b>"] --> SDD["<b>subagent-driven-development</b><br/>coordinator"]
    SDD --> Worktrees["<b>using-git-worktrees</b><br/>.worktrees/<task-slug>/"]
    Worktrees --> Swarm["<b>dispatching-parallel-agents</b><br/>one AgentSwarm call"]
    
    subgraph Wave1["Wave 1"]
        WT1["worktree/ticket-1"] --> SA1["subagent<br/><b>conventional-commits</b>"] --> Squash1["squash → feat branch"]
    end
    
    subgraph Wave2["Wave 2 (parallel)"]
        WT2["worktree/ticket-2"] --> SA2["subagent<br/><b>conventional-commits</b>"] --> Squash2["squash → feat branch"]
        WT3["worktree/ticket-3"] --> SA3["subagent<br/><b>conventional-commits</b>"] --> Squash3["squash → feat branch"]
    end
    
    Swarm --> Wave1
    Swarm --> Wave2
    
    Squash1 --> Integrate["Integrate<br/>(squash + commit)<br/>conflicts → <b>resolving-merge-conflicts</b>"]
    Squash2 --> Integrate
    Squash3 --> Integrate
    
    Integrate --> Docs["<b>forge-docs</b><br/>update indexes"]
    Docs --> Verify["<b>verification-before-completion</b><br/>full test suite"]
    
    style Plan fill:#27AE60,color:#fff,stroke:#1A7A42
    style SDD fill:#16A085,color:#fff,stroke:#0E6655
    style Worktrees fill:#F39C12,color:#fff,stroke:#B8750E
    style Swarm fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style WT1 fill:#F39C12,color:#fff,stroke:#B8750E
    style WT2 fill:#F39C12,color:#fff,stroke:#B8750E
    style WT3 fill:#F39C12,color:#fff,stroke:#B8750E
    style SA1 fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style SA2 fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style SA3 fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style Squash1 fill:#16A085,color:#fff,stroke:#0E6655
    style Squash2 fill:#16A085,color:#fff,stroke:#0E6655
    style Squash3 fill:#16A085,color:#fff,stroke:#0E6655
    style Integrate fill:#16A085,color:#fff,stroke:#0E6655
    style Docs fill:#E67E22,color:#fff,stroke:#A05A15
    style Verify fill:#27AE60,color:#fff,stroke:#1A7A42

5. Review — PR Draft → Review → Resolve Loop

After squash commits land, creating-pull-requests drafts the PR with ste100 prose (draft mode, AI disclosure). verification-before-completion re-runs the full suite on the merged feat branch. pr-review runs in a worktree, drives a kimi -p loop via loops, and posts ONE review under the authenticated user's identity in caveman-review format. If findings come back 🔴 or 🟡, pr-resolve consumes them: one resolver per finding group, each in its own worktree off the PR head, commits under use-git-identity, pushes, replies in each review thread, and re-runs pr-review. The loop ends when no 🔴/🟡 findings remain.

flowchart LR
    Verify["<b>verification-before-completion</b><br/>suite green"] --> PR["<b>creating-pull-requests</b><br/>PR draft (<b>ste100</b> prose)"]
    PR --> Review["<b>pr-review</b><br/><b>loops</b> + <b>caveman-review</b>"]
    Review --> Findings{"🔴/🟡<br/>findings?"}
    Findings -->|"yes"| Resolve["<b>pr-resolve</b><br/>per-group worktrees<br/><b>use-git-identity</b>"]
    Resolve --> Push["Push + Reply<br/>in review threads"]
    Push --> ReReview["<b>pr-review</b><br/>re-run"]
    ReReview --> Findings
    Findings -->|"no"| Merge(["User merges"])
    
    style Verify fill:#27AE60,color:#fff,stroke:#1A7A42
    style PR fill:#2C3E50,color:#fff,stroke:#1A252F
    style Review fill:#E67E22,color:#fff,stroke:#A05A15
    style Findings fill:#F39C12,color:#fff,stroke:#B8750E
    style Resolve fill:#E67E22,color:#fff,stroke:#A05A15
    style Push fill:#16A085,color:#fff,stroke:#0E6655
    style ReReview fill:#E67E22,color:#fff,stroke:#A05A15
    style Merge fill:#95A5A6,color:#fff,stroke:#7F8C8D

6. File Artefacts Per Step

Each forge step emits concrete files. The table below shows what gets written where.

Step	Artefact	Location
1. Map	Labels `wayfinder:*`	GitHub labels
1. Map	Map + ticket issues	GitHub tracker
2. Resolve	Glossary terms	`docs/dev/CONTEXT.md`
2. Resolve	ADRs	`docs/adr/NNNN-*.md`
2. Resolve	Research reports	`docs/dev/agents/<topic>.md`
3. Plan	Plan file	plan mode output
4. Work	Worktrees	`.worktrees/<task-slug>/`
4. Work	Feat branch	`feat/wayfinder-<map>`
5. PR	PR draft	`gh api`
6. Verify	Full-suite run output	terminal
7. Review	Inline comments	PR review threads
8. Resolve	Resolver worktrees	`.worktrees/<finding>/`

Skills

Skill	Purpose	Load	Source	Guide
`skills/forge/SKILL.md`	Session-start orchestrator: map → resolve → plan → work → verify → review → resolve	always	local	`docs/guides/forge.md`
`skills/forge-flow/SKILL.md`	Session bootstrap: feat branch from main, harness goal, hand off to forge step 1	always	local	`docs/guides/forge-flow.md`
`skills/forge-init/SKILL.md`	Bootstrap a repo to be forge-ready: AGENTS.md contract, grilling for repo-specifics	one-shot	local	`docs/guides/forge-init.md`
`skills/forge-setup/SKILL.md`	One-shot harness adaptation: discover non-agnostic references, research the running harness's equivalent, patch with minimal diffs	one-shot	local	`docs/guides/forge-setup.md`
`skills/forge-docs/SKILL.md`	Maintain the docs directory — structure, update rules, index files, ADR mandate	on-demand	local	`docs/guides/forge-docs.md`
`skills/forge-cleanup/SKILL.md`	Remove stale forge artefacts — scratch files, worktrees, uncommitted changes, local branches	one-shot	local	`docs/guides/forge-cleanup.md`
`skills/wayfinder/SKILL.md`	Plan a huge effort as a shared map of decision tickets on the issue tracker	always	mattpocock/skills	`docs/guides/wayfinder.md`
`skills/caveman/SKILL.md`	Ultra-compressed chat replies — drop articles/filler, keep technical accuracy	always	JuliusBrussee/caveman	`docs/guides/caveman.md`
`skills/grilling/SKILL.md`	Grill the user relentlessly — design-tree frontier, max 4 questions per wave	on-demand	mattpocock/skills	`docs/guides/grilling.md`
`skills/prototype/SKILL.md`	Build a throwaway prototype to answer a design question	on-demand	mattpocock/skills	`docs/guides/prototype.md`
`skills/deep-research/SKILL.md`	Exhaustive evidence-based research — 10+ iteration search loop, markdown-native reports	on-demand	MoweME (origin: kimi.com web UI)	`docs/guides/deep-research.md`
`skills/domain-modeling/SKILL.md`	Build and sharpen the domain model: CONTEXT.md glossary + ADRs in `docs/adr/`	on-demand	mattpocock/skills	`docs/guides/domain-modeling.md`
`skills/git-issue-tracker/SKILL.md`	Wrap the GitHub API for issue/sub-issue/dependency operations used by wayfinder	on-demand	local	`docs/guides/git-issue-tracker.md`
`skills/planning-and-task-breakdown/SKILL.md`	Break a spec into ordered, implementable tasks with parallel work identified	on-demand	local	`docs/guides/planning-and-task-breakdown.md`
`skills/using-git-worktrees/SKILL.md`	Create one worktree per task, commits inside, cleanup by coordinator	on-demand	obra/superpowers	`docs/guides/using-git-worktrees.md`
`skills/dispatching-parallel-agents/SKILL.md`	AgentSwarm mechanics: dispatch up to 10 parallel subagents via `{{item}}` prompt template	on-demand	obra/superpowers	`docs/guides/dispatching-parallel-agents.md`
`skills/subagent-driven-development/SKILL.md`	Coordinator dispatch → per-task review → fix loop → integrate. One worktree per task	on-demand	obra/superpowers	`docs/guides/subagent-driven-development.md`
`skills/finishing-a-development-branch/SKILL.md`	Push + PR creation path; merge conflicts load `resolving-merge-conflicts`	on-demand	obra/superpowers	`docs/guides/finishing-a-development-branch.md`
`skills/verification-before-completion/SKILL.md`	Iron Law: NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE	on-demand	obra/superpowers	`docs/guides/verification-before-completion.md`
`skills/pr-review/SKILL.md`	Validate → review loop → post ONE review (caveman-review findings) inside a worktree	on-demand	local	`docs/guides/pr-review.md`
`skills/pr-resolve/SKILL.md`	Findings → fix → push → thread replies (loop until no 🔴/🟡)	on-demand	local	`docs/guides/pr-resolve.md`
`skills/loops/SKILL.md`	Shell framework: render prompt template, cavemanize, drive `kimi -p` until DONE:/BLOCKED:	on-demand	local	`docs/guides/loops.md`
`skills/creating-pull-requests/SKILL.md`	Size-gated PR descriptions with mandatory Conventional Commit title and AI disclosure (prose per `ste100`)	on-demand	tdhopper/dotfiles2.0	`docs/guides/creating-pull-requests.md`
`skills/caveman-review/SKILL.md`	Ultra-compressed code review comments: location, problem, fix — one line per finding	on-demand	JuliusBrussee/caveman	`docs/guides/caveman-review.md`
`skills/ste100/SKILL.md`	Write human-facing text in ASD-STE100 Simplified Technical English	on-demand	local	`docs/guides/ste100.md`
`skills/caveman-commit/SKILL.md`	Ultra-compressed commit messages in Conventional Commits format	on-demand	JuliusBrussee/caveman	`docs/guides/caveman-commit.md`
`skills/conventional-commits/SKILL.md`	Conventional Commits v1.0.0 spec — type→SemVer table	on-demand	local	`docs/guides/conventional-commits.md`
`skills/use-git-identity/SKILL.md`	Set git identity (weselben/rooforge) before any commit/amend/rebase	on-demand	local	`docs/guides/use-git-identity.md`
`skills/resolving-merge-conflicts/SKILL.md`	Resolve git merge/rebase conflicts; multi-branch conflicts delegate to SDD	on-demand	mattpocock/skills	`docs/guides/resolving-merge-conflicts.md`
`skills/12-factor-app/SKILL.md`	SaaS / cloud-native design and review against the 12-Factor methodology and modern extensions	on-demand	12factor.net	`docs/guides/12-factor-app.md`
`skills/kiss-principle/SKILL.md`	Simplicity guardrails — over-engineering, premature abstraction, accidental complexity	on-demand	Wikipedia	`docs/guides/kiss-principle.md`
`skills/frontend-design/SKILL.md`	Distinctive, intentional visual design — typography, palette, layout, anti-templated defaults	on-demand	anthropics/skills	`docs/guides/frontend-design.md`
`skills/forge-tailwindcss-conventions/SKILL.md`	Tailwind CSS v4 conventions — CSS-first, utility-first, class ordering, framework patterns	on-demand	local	`docs/guides/forge-tailwindcss-conventions.md`
`skills/forge-eu-accessibility/SKILL.md`	EU accessibility compliance — BFSG/EAA scope, WCAG 2.1 AA baseline, penalties, live verification	on-demand	local	`docs/guides/forge-eu-accessibility.md`
`skills/forge-seo/SKILL.md`	SEO hub — UX/UI and Technical deep-dives (Core Web Vitals, sitemaps, structured data, rendering)	on-demand	local	`docs/guides/forge-seo.md`

Repository Structure

.
├── docs/
│   ├── README.md           # Global docs index
│   ├── adr/                # Architecture Decision Records
│   ├── dev/
│   │   ├── README.md       # dev subfolder index
│   │   ├── CONTEXT.md      # Domain glossary (ADR index)
│   │   └── agents/         # Deep research reports
│   ├── guides/
│   │   ├── README.md       # guides subfolder index
│   │   └── <skill>.md      # One reference guide per skill (35 total)
│   ├── public/
│   │   └── README.md       # public subfolder index
│   └── system-design/
│       └── README.md       # system-design subfolder index
├── skills/
│   ├── 12-factor-app/      # sidecar: SaaS / cloud-native conventions
│   ├── caveman/
│   ├── caveman-commit/
│   ├── caveman-review/
│   ├── conventional-commits/
│   ├── creating-pull-requests/
│   ├── deep-research/
│   ├── dispatching-parallel-agents/
│   ├── domain-modeling/
│   ├── finishing-a-development-branch/
│   ├── forge/                # the orchestrator
│   ├── forge-cleanup/
│   ├── forge-docs/
│   ├── forge-eu-accessibility/   # sidecar: BFSG/EAA/WCAG compliance
│   ├── forge-flow/
│   ├── forge-init/
│   ├── forge-seo/               # sidecar: SEO hub
│   ├── forge-setup/             # one-shot harness adaptation
│   ├── forge-tailwindcss-conventions/  # sidecar: Tailwind v4 conventions
│   ├── frontend-design/         # sidecar: distinctive visual design
│   ├── git-issue-tracker/
│   ├── grilling/
│   ├── kiss-principle/          # sidecar: simplicity guardrails
│   ├── loops/
│   ├── planning-and-task-breakdown/
│   ├── pr-resolve/
│   ├── pr-review/
│   ├── prototype/
│   ├── resolving-merge-conflicts/
│   ├── ste100/
│   ├── subagent-driven-development/
│   ├── use-git-identity/
│   ├── using-git-worktrees/
│   ├── verification-before-completion/
│   └── wayfinder/
├── src/                  # reserved
├── tests/                # reserved
├── AGENTS.md             # Local agent contract (letter style)
└── README.md             # this file

Docs

Every skill ships with a one-page reference guide under docs/guides/ — the index links all 35. The global docs index at docs/README.md covers the rest: docs/dev/CONTEXT.md (domain glossary + ADR index), docs/adr/ (architecture decisions), docs/system-design/, docs/public/, and docs/dev/agents/ (deep research reports).

A local agent contract lives at AGENTS.md — letter style, scoped to this repo's conventions.

Invariant Rules

These rules bind every session where forge is active:

Mandates are mandatory. When any skill uses must, mandatory, MUST, or always, follow it exactly.
caveman is default. caveman(ultra) is active every response. Off only on explicit "stop caveman" / "normal mode".
Own repos only. Forge operates only in repos owned by the user's git identity.
Single path. One flow through the steps. No branching in the orchestrator.
Forge Flow first. forge-flow runs before step 1: creates the feat branch from main, sets the harness goal, hands off.

Navigation Menu

Repository files navigation

RooForge

Overview

Token budget

Install

Prerequisites

The Flow

1. Session Start — Forge loads, forge-flow bootstraps

2. Map — Load or Chart the Wayfinder Map

3. Resolve — Work One Ticket

4. Work — Subagent-Driven Development

5. Review — PR Draft → Review → Resolve Loop

6. File Artefacts Per Step

Skills

Repository Structure

Docs

Invariant Rules

About

Releases

Contributors

Languages

Name		Name	Last commit message	Last commit date
.github		.github
docs		docs
mcp		mcp
skills		skills
src		src
tests		tests
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
mcp.md		mcp.md

Latest commit

History

Folders and files

Repository files navigation

RooForge

Overview

Token budget

Install

Prerequisites

The Flow

1. Session Start — Forge loads, forge-flow bootstraps

2. Map — Load or Chart the Wayfinder Map

3. Resolve — Work One Ticket

4. Work — Subagent-Driven Development

5. Review — PR Draft → Review → Resolve Loop

6. File Artefacts Per Step

Skills

Repository Structure

Docs

Invariant Rules

About

Topics

Resources

Contributing

Stars

Watchers

Forks

Releases

Contributors

Languages