codetruss — Claude Code plugin: what it does, how to install
The codetruss plugin for Claude Code and Codex sets agent file boundaries, runs local analyzers on code changes, and creates signed review receipts.
MIT ★ 1,596 updated 2026-10-29
/plugin marketplace add hashgraph-online/awesome-codex-plugins
/plugin install codetruss@awesome-codex-pluginsWhat codetruss does
This plugin connects an agent session to the CodeTruss CLI. It defines allow and deny file perimeters for each task so the agent cannot silently modify sensitive files or policies. After edits finish, the CLI tests the result against the repository verification commands on an immutable Git tree.
The local security pass targets JavaScript, TypeScript, and TSX files. It checks for SQL injection, mass assignment, un-awaited database writes, and N+1 query loops. Every finished review produces a signed receipt that you can validate offline with codetruss verify.
The plugin acts as a thin wrapper and will not run without a pre-installed CodeTruss CLI v0.2.35 or newer and Node.js 20.9+.
What it helps with
- stop coding agents from touching files outside designated scope
- catch hardcoded credentials before opening a pull request
- detect common database mistakes and SQL injections in TypeScript
- generate an offline-verifiable audit receipt for generated commits
Example prompts for the agent
- Set up CodeTruss for this repository
- Check scope boundaries for recent edits
- Show the latest CodeTruss verification report
How to install codetruss
Plugin marketplace
/plugin marketplace add hashgraph-online/awesome-codex-plugins/plugin install codetruss@awesome-codex-pluginsStep by step, with commands for each platform: Claude Code plugins: marketplaces, installing, and building your own
Questions about codetruss
How to install codetruss in Claude Code?
Register the marketplace with /plugin marketplace add hashgraph-online/awesome-codex-plugins, then run /plugin install codetruss@awesome-codex-plugins.
What does codetruss do?
It enforces strict file boundaries, runs local deterministic analyzers, and writes signed verification receipts for agent changes.
What dependencies does codetruss need?
You need Git, Node.js 20.9 or newer, and CodeTruss CLI v0.2.35 or newer installed on your machine.
Source text
codetruss-plugins · MIT
# CodeTruss agent plugins CodeTruss is the deterministic first-pass verification gate for AI-written code. This repository is the open wrapper that puts that gate inside Claude Code and Codex: a skill that teaches the agent to operate the separately installed CodeTruss CLI, bind each task to an approved file boundary, run the repository's own verification commands, and read the signed receipt the CLI writes for every change. The analysis is deterministic and runs on the developer's machine, the receipt records exactly what ran and what did not, and anyone can re-verify a receipt offline. Nothing here calls a model, uploads source, or scores a repo by vibes. The wrappers contain no analyzer, no bundled hook, no MCP server, and no upload path. They invoke the separately installed CLI, keep deterministic analysis local, and require explicit consent before installation, provider-backed `--llm` review, authentication, or receipt `sync`. ## Install for Claude Code Inside a Claude Code session: ```text /plugin marketplace add DeliriumPulse/codetruss-plugins /plugin install codetruss@codetruss ``` Or from a shell: ```bash claude plugin marketplace add DeliriumPulse/codetruss-plugins claude plugin install codetruss@codetruss ``` Then ask Claude to `Set up CodeTruss for this repository`. The skill proposes an allow/deny boundary and verification commands, waits for your confirmation, and runs the CLI's own guided `codetruss setup` with `--hooks claude` so the tested CLI installer owns hook installation. From then on the gate runs on the agent's lifecycle events and writes a receipt per session. The owned marketplace is live now. CodeTruss is not currently listed in Anthropic's reviewed community catalog or its separately curated official marketplace. ## Prerequisites - Git - Node.js 20.9 or newer - CodeTruss CLI v0.2.35 or newer from <https://codetruss.com/cli> Skill instructions in this repository are verified against the shipped CLI v0.2.35. The skill can explain the official installer when the CLI is missing, but it must not install software without the developer's confirmation. ## What a receipt looks like Verbatim excerpt from `codetruss report latest` after a review where the agent also touched a denied path and committed a live key: ```markdown # CodeTruss receipt — FAILED - **Session:** `20260807T051345990Z-4f2f16` - **Task:** Add a SESSION_TTL constant to src/auth.ts - **Starting commit:** `14482ad37ea1f3ea5c9547071b58c3cdffb7c0b0` - **Policy SHA-256:** `d13f4367129bb7683c486d4d4e8e876bc1e9a7f541cd0e539945743b16d95a03` ## Verdict: FAILED - 1 high/critical security or dependency finding(s) affect changed files - 1 file(s) changed in denied paths: infra/deploy.sh - 1 file(s) changed outside approved scope: .codetruss.yml - sensitive surfaces changed: .codetruss.yml (policy) ## Changed files (3) | Path | Change | Scope | Sensitive | Lines | |---|---|---|---|---:| | `.codetruss.yml` | added | unexpected | policy | +5/−0 | | `infra/deploy.sh` | added | denied | — | +3/−0 | | `src/auth.ts` | modified | allowed | — | +1/−0 | ## Introduced or worsened analyzer findings (1) | Severity | Analyzer | Location | Finding | |---|---|---|---| | HIGH | secrets | `infra/deploy.sh:2` | Possible Stripe live secret key committed in deploy.sh | ## Analysis profile Profile: `local-registry-v2`. The 13 deterministic registry analyzers ran locally on this machine, plus a local security pass: the shared SAST engine — the same rules and the same source-to-sink taint tracking as the hosted audit — over the JavaScript, TypeScript and TSX in this repository. ### What the local security pass checked - **SQL injection (CWE-89).** Untrusted input tracked from request sources through string building into query execution. - **Mass assignment (CWE-915).** A raw request body spread into a database write, and write helpers whose payload type accepts arbitrary keys. - **Un-awaited database writes, swallowed errors, coercion-prone `==` comparisons, and N+1 queries in loops** — the defect classes coding agents most often introduce. ### What did not run - **The rest of the security rule pack.** Command injection, code injection, path traversal, SSRF, open redirect, XSS and insecure deserialization were **not** checked here. Those rules run in a hosted scan; absence of a finding in those classes means they were not analyzed, not that the code is clean. - **Non-JavaScript languages.** The local pass covers JavaScript, TypeScript and TSX only. Python, Go, Java, C#, PHP, Ruby and Rust in this repository received secret scanning and the other registry passes, but no security rule or taint analysis. - **Hosted symbol graph.** No cross-file call or data-flow graph was built, so architecture and dead-code conclusions cover only what the local passes can see in isolation. - **Optional LLM review.** No model read this diff. It is opt-in via `--llm` and is force-disabled under agent hooks, so a hook receipt is always deterministic evidence only. - **Hosted Health scores.** Not calculated, reported as **N/A**. The scores are defined over the graph and the complete SAST pass; a number derived from this pass set would overstate what ran. Local security findings are reported for review and do not fail the verdict on their own. ``` `codetruss verify latest` re-checks the signature over those bytes and prints `verified 20260807T051345990Z-4f2f16 (FAILED)`. ## What runs where Local and deterministic, with no account and no network required: - the 13 registry analyzers listed in the receipt's analysis profile; - since CLI v0.2.35, a security pass running the shared SAST engine and its source-to-sink taint tracking over the repository's JavaScript, TypeScript and TSX: SQL injection, mass assignment, un-awaited database writes, swallowed errors, coercion-prone `==`, and N+1 queries in loops; - scope classification of every changed file against the approved allow/deny boundary, including untracked and sensitive-surface detection; - secret detection on changed files; - the repository's own verification commands, executed against a fresh materialization of the final Git tree; - receipt signing and offline `codetruss verify`. Hosted, and only when the developer explicitly asks for it: - the rest of the security rule pack — command injection, code injection, path traversal, SSRF, open redirect, XSS, insecure deserialization — and every non-JavaScript language; - the cross-file symbol graph, which together with the full rule pack backs the full-codebase audit and the Health scores; - receipt `sync` to a CodeTruss account. A local receipt names its own remaining gaps in its "What did not run" section, and local security findings are `REVIEW_REQUIRED` at most — they never fail a verdict on their own. Optional provider-backed `--llm` review is force-disabled under agent hooks, so a hook receipt is always deterministic evidence only. Measured behaviour on public repositories is published at <https://codetruss.com/benchmark>. CLI downloads, checksums, and the SBOM are at <https://codetruss.com/cli>. ## Codex ```bash codex plugin marketplace add DeliriumPulse/codetruss-plugins codex plugin add codetruss@codetruss ``` The owned marketplace is live now. CodeTruss is not currently listed in OpenAI's public Plugin Directory. ## Agent Skills clients Install the same canonical skill for both Claude Code and Codex with the public Agent Skills installer: ```bash npx --yes skills add DeliriumPulse/codetruss-plugins \ --skill codetruss --agent claude-code codex -y ``` For an immutable install, pin the most recent published release: ```bash npx --yes skills add \ https://github.com/DeliriumPulse/codetruss-plugins/tree/v0.1.5 \ --skill codetruss --agent claude-code codex -y ``` The skill is indexed at <https://skills.sh/DeliriumPulse/codetruss-plugins/codetruss>. Review the skill before use; it can invoke the separately installed CLI with the agent's local permissions. ## What the skill does - proposes a narrow repository allow/deny policy and verification commands; - runs guided `codetruss setup` only after the developer confirms the boundary, hook target, and exact verification-command trust; - installs and diagnoses the existing Claude Code or Codex lifecycle hook; - reviews working-tree or staged changes and interprets the resulting receipt; - treats exit codes 1 and 2 as product verdicts with valid evidence, not generic shell failures; - refuses to weaken policy merely to manufacture a green verdict. The canonical Agent Skills definition is in `skills/codetruss/`. Platform packages contain byte-identical copies so each marketplace archive is self-contained. ## Development ```bash npm test claude plugin validate . --strict claude plugin validate ./plugins/codetruss-claude --strict python3 /path/to/plugin-creator/scripts/validate_plugin.py ./plugins/codetruss npm run release:verify ``` `npm test` uses only Node.js built-ins and verifies manifests, marketplace entries, skill parity, and privacy guardrails. `npm run release:verify` builds the OpenAI and Claude plugin archives twice from the committed Git tree and proves their bytes are reproducible. ## Security and support Read [SECURITY.md](SECURITY.md) before reporting a vulnerability. General wrapper issues belong in this repository; CLI issues belong in the [CodeTruss CLI tracker](https://github.com/DeliriumPulse/codetruss-cli/issues). Never attach third-party source, credentials, or unredacted receipts to a public issue. ## Licensing boundary The manifests and skill instructions in this repository are MIT licensed. The separately installed CodeTruss CLI is free to use under the CodeTruss CLI Proprietary License and CodeTruss Terms of Service. This repository does not make the CLI open source and does not redistribute its executable bundle.
© CodeTruss · MIT · hashgraph-online/awesome-codex-plugins