Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/ariegoldkin/claude-forge/claude-mdgit clone --depth 1 https://github.com/ArieGoldkin/claude-forgeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/ariegoldkin/claude-forge/claude-md)<a href="https://agentmods.dev/instructions/ariegoldkin/claude-forge/claude-md"><img src="https://agentmods.dev/badge/instructions/ariegoldkin/claude-forge/claude-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.13964 | $0.13964 |
| Opus 5 | $0.06982 | $0.06982 |
| Sonnet 5 | $0.02793 | $0.02793 |
| Haiku 4.5 | $0.01396 | $0.01396 |
Grade F, and why
claude-forge CLAUDE.md scanned grade F with 6 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Sends data to an external URLmediumData exfiltration
A POST to an outside endpoint may be telemetry or may be exfiltration; either way the mod talks to somewhere, and you should know where.
> **CC Alignment**: **v2.1.221 (triaged 2026-08-04)** — `docs/reviews/2026-08-04_cc-v2.1.221-alignment-audit.md`. That pass closed the 28-version gap by **triage**, not by behavioural probing: it measured what this repo Reaches for credential fileshighPrivilege escalation
SSH keys, cloud credentials, git-credentials, .npmrc, /etc/shadow: reading these is how a config file becomes a credential leak.
> **CC Alignment**: **v2.1.221 (triaged 2026-08-04)** — `docs/reviews/2026-08-04_cc-v2.1.221-alignment-audit.md`. That pass closed the 28-version gap by **triage**, not by behavioural probing: it measured what this repo Downloads and executes remote codehighSupply chain
curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.
"Bash(curl * | sh*)", Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
Why this matters here: **auto-research is the repo's highest-fan-out entry point** (a `design` route spawns ~11 agents), so it is exactly where an unconstrained model choice is most expensive. These settings are the enfo Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
"Bash(rm -rf /*)", Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
> **CC Alignment**: **v2.1.221 (triaged 2026-08-04)** — `docs/reviews/2026-08-04_cc-v2.1.221-alignment-audit.md`. That pass closed the 28-version gap by **triage**, not by behavioural probing: it measured what this repo How it starts
The opening of the file, as written. The whole thing — 605 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Claude Forge
Maintainer: Arie Goldkin Plugins: ctk (formerly continuity-toolkit), dtk (formerly devops-toolkit), atk (formerly ai-toolkit), ftk (formerly frontend-toolkit), etk (formerly engineering-toolkit) Versioning: Independent per plugin CC Alignment: v2.1.221 (triaged 2026-08-04) —
docs/reviews/2026-08-04_cc-v2.1.221-alignment-audit.md. That pass closed the 28-version gap by triage, not by behavioural probing: it measured what this repo ships and declares against the v2.1.194→221 CHANGELOG, and fixed the two claims found stale — the subagent nesting depth (5 → 3, having passed through 1) andcontext: forkskills now running in the background by default. ⚠ Its Tier-2/3 items are explicitly labelled hypotheses, not findings, because the previous audit's false floor note came from treating a CHANGELOG line as a conclusion. The auto-mode settings-path question (v2.1.207) is the one most likely to bite and is deliberately left unresolved pending a probe. Prior: v2.1.193 (audited 2026-06-26); the v2.1.174→193 gap analysis (model-governance enforcement viaenforceAvailableModels, the hookif-condition file-path fix, auto-mode native git-blocking +classifyAllShell, nested-subagent 5-deep (superseded — the live default is 3, see the Subagent section),/pluginSkills tab) is indocs/reviews/2026-06-26_cc-v2.1.193-alignment-gaps.md. Floor note — RETRACTED 2026-08-01, it was wrong (issue #98). It previously read that ctk'shooks.jsonfile-pathifconditions depend on the v2.1.176 fix "to gate correctly," making that the effective minimum CC version for our security/permission hooks. Measured on v2.1.220: 4 of the 6 conditions were probed directly and none gated — neither the Bash ones nor the file-path ones (the other 2 are inferred, not measured — see the write-up). All 6 were removed from ctk'shooks.json; each hook's own internal guard is and always was the real filter. Cause (best-supported explanation, from the official hooks docs plus measurement): they were wrong twice over —ifis a hook-handler field (sibling oftype/command), not a matcher-group field, so at group level the key was never read; andif"holds exactly one permission rule — there is no&&,||, or list syntax," while every value we shipped was a||-joined list. ⚠ The correct-looking fix is the dangerous one — do NOT "fix the nesting." Movingifinto the handlers and splitting the lists is exactly what the CC docs prescribe in general, and onPreToolUse[0]it would confinesecurity-blockerto 20 destructive-verb globs, letting every secret-file read through (cat <envfile>,curl -d @<envfile>,cat ~/.ssh/id_rsamatch none of them).PreToolUse[2]is the same trap for the write path. CC's own docs agree on the principle: "theiffilter is best-effort, use the permission system rather than a hook to enforce a hard allow or deny." Pinned byplugins/continuity-toolkit/hooks/tests/hooks-manifest.test.ts. The false claim originated in a CHANGELOG gap-analysis that verified the conditions were present and never that they gate. Full measurements, the six-key landmine table, and what remains NOT ESTABLISHED (what CC does with a malformedifat the correct nesting level — undocumented; and the "never read at group level" step is an inference from measurement, not a documented fact):docs/reviews/2026-08-01_98-hook-if-condition-cause.md. Per-version "considered but not adopted" notes are not tracked here — seegit log -p CLAUDE.mdor the CC CHANGELOG for prior reasoning. ⚠ Two earlier reviews are cited by this header but have no write-up in the repo: the v2.1.159→173 and v2.1.153→158 passes nameddocs/reviews/2026-06-11_cc-v2.1.173-alignment-audit.mdanddocs/reviews/2026-05-31_cc-v2.1.158-plugin-system-alignment-audit.md. Measured 2026-08-04: neither file has ever been committed on any branch (git log --all --diff-filter=Areturns 0 for both), so those citations have always been dangling. Their findings survive only in this header and ingit log -p CLAUDE.md. Do not go looking for the files.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 605 lines · 13,964 tokens per session scan F 0f1982f391fc
claude-forge CLAUDE.md is an instructions file published in the GitHub repository ArieGoldkin/claude-forge (6 stars, last pushed 28d ago), licensed MIT. It adds 13,964 tokens to every session, about $0.0698 per session on Opus 5. A static security scan graded it F with 6 findings (sends data to an external url, reaches for credential files, downloads and executes remote code). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
agent-plugins codeowners.instructions.md
Instructions for awslabs/agent-plugins, covering codeowners file rules, plugins (alphabetically listed) and file must end with codeowners file.
agent-plugins copilot-instructions.md
Copilot instructions for awslabs/agent-plugins, covering project overview, technology stack, code organization, development practices and plugin development guidelines.
agent-plugins CLAUDE.md
Claude Code instructions for awslabs/agent-plugins, a project described as: Agent Plugins for AWS equip AI coding agents with the skills to help you architect, deploy, and operate on AWS.
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
spec-kit AGENTS.md
AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.