Skip to content
Skills Library
Programming · Review & audit

forge

/forge

Use when a non-trivial software change needs independent cross-model challenge or bounded multi-unit implementation.

Repository
forge ↗
Licence
MIT
Last repository change
28.10.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
12
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 28.10.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Use when a non-trivial software change needs independent cross-model challenge or bounded multi-unit implementation.

Original author description

Use when a non-trivial software change needs independent cross-model challenge or bounded multi-unit implementation.

How it works

  1. Start here
  2. Profiles
  3. Non-negotiable authority
  4. Load detail only when needed
  5. Stop conditions
Original SKILL.md headings · 70 lines, 502 words
SKILL.md preview
---
name: forge
description: Use when a non-trivial software change needs independent cross-model challenge or bounded multi-unit implementation.
---

# Forge

Forge is the shared authority for reviewed Codex-and-Claude builds. Codex remains
the driver and integration authority; external Claude work is bounded by typed bridge
tools, immutable work orders, admission, receipts, and the selected execution profile.

## Start here

1. Read the role and control boundary in [ENGINE.md](ENGINE.md).
2. Resolve the profile in this order: explicit owner instruction, repository
   `.ai/EXECUTION_PROFILE.md`, then `assured`.
3. Read [ROUTING-POLICY.md](ROUTING-POLICY.md) before assigning any unit.
4. Complete [Acts 1–2](references/acts-1-2.md). Neither act writes product code.
5. After Gate 1 is genuinely approved, select the narrow Act 3 or dependency-aware
   Act 4 procedure in [Acts 3–4](references/acts-3-4.md).

Owner instructions and repository policy always bind. Forge adds no authority beyond
the current task, repository, or environment.

## Profiles

| Profile | Cadence authority |
|---|---|
| [Sprint](profiles/sprint.md) | Smallest relevant proof per unit, immediate serial integration, one comprehensive sprint-end gate. |
| [Assured](profiles/assured.md) | Genuine cross-model receipt per unit, Codex complete-diff review and proof rerun, integrated gate. |

Gate 1 applies to both profiles. Gate 2 exists only for Assured complex work. A red
Sprint proof is fixed and rerun at the same scope; it never promotes the profile by
itself. There is no invented owner commit-approval gate.

## Non-negotiable authority

- Codex conducts the grill, decides what enters the plan, arbitrates routes and
  findings, verifies results, integrates, and alone may commit, merge, or push after
  owner and repository gates pass.
- Fable 5 at max is the default external reviewer and executor. Sol 5.6 at ultra is
  the default Codex-side route. Either model may own a routed unit and use native
  subagents.
- Same-model review is never called cross-model. The executor never commits, merges,
  or pushes.
- The bridge controls Claude only. Codex-native lanes use native Codex subagents.
- Never control Claude outside the [typed bridge contract](references/bridge-tools.md).
- Let the trusted plugin hook bind the active task workspace automatically. Never
  create or modify the reserved `_forge_workspace` tool input.
- No lane executes until executable admission returns `ADMIT`. At most three lanes
  may be active, with one machine-wide heavy lease, one writer per worktree, and
  disjoint parallel file sets.

## Load detail only when needed

| Need | Authority |
|---|---|
| Roles, model defaults, integration boundary | [Engine](ENGINE.md) |
| Profile-sensitive assignment and gates | [Routing policy](ROUTING-POLICY.md) |
| Grill and adversarial plan review | [Acts 1–2](references/acts-1-2.md) |
| Frozen and orchestrated execution | [Acts 3–4](references/acts-3-4.md) |
| Exact MCP surface for all Claude control | [Bridge tools](references/bridge-tools.md) |
| Executable guard and atomic reservation | [Admission](references/admission.md) |
| Owner budgets and provider pauses | [Usage limits](references/usage-limits.md) |
| Objective progress and recovery observation | [Liveness](references/liveness.md) |

## Stop conditions

Stop rather than improvise when Gate 1 has no approval, admission rejects, the owner
budget reaches its cutoff, a provider/policy pause occurs, the baseline or allowed
files drift, a required receipt is missing, or a requested route would require an
unapproved model substitution. Preserve the plan, worktree, logs, job and revision,
then ask for the decision that is actually missing.
Files in this package (12)
  • ENGINE.md 5 KB
  • ROUTING-POLICY.md 4 KB
  • SKILL.md 4 KB
  • agents/openai.yaml 1 KB
  • profiles/assured.md 3 KB
  • profiles/sprint.md 3 KB
  • references/acts-1-2.md 7 KB
  • references/acts-3-4.md 19 KB
  • references/admission.md 12 KB
  • references/bridge-tools.md 9 KB
  • references/liveness.md 5 KB
  • references/usage-limits.md 5 KB

Versions

RepositoryScoreLicence
This page · 3orionkings/forge90MIT
aimi-so/aimi-engineering-plugin100MIT
claudemods/context-forge100MIT
cwinvestments/memstack100MIT
MohammedAtya/guardrail-forge100MIT
faizanio/FORGE100MIT
thecrazybob/claude-code-plugins100MIT
floka-as/floka-marketplace100MIT

Database snapshot: 13.11.2026 · package version 0e1786211440 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 90forge-with-docs
    Use when planning a non-trivial change in a repository whose domain glossary, context model, or ADRs must remain authoritative.
  • 96forge-build
    Use when an approved frozen work order defines one bounded implementation unit.
  • 100next-forge
    Expert assistance for next-forge — a production-grade Turborepo template for Next.js SaaS apps. Triggers on questions about next-forge installation, setup, architecture, packages, customization, deployment, and development workflows.
  • 100forge-cli
    Use when an agent needs to call a running three.ws Forge desktop instance from the terminal to generate/export image-to-3D assets through canonical JSON-first automation commands.
  • 100forge-design
    Tyler Forge design system expert. Trigger when the user mentions Forge, Forge components, Forge blocks, @tylertech/forge, forge-tailwind, tyler-icons, forge-scaffold, forge-card, forge-app-bar, forge-table, forge-dialog, forge-drawer, forge-list, forge-field,…
  • 100forge-guard
    Drive an open pull request to review-ready without a human watching — run simplification and security passes, wait for CI and the peer reviewer, address feedback in bounded rounds, and stop at review-ready or needs-human. Never merges. Use when the user wants to walk away from a…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.