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 skills/a16z/jolt/ci-code-reviewnpx skills add a16z/jolt --skill ci-code-reviewgit clone --depth 1 https://github.com/a16z/joltWhat 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 | $0.00129 | $0.01840 |
| Opus 5 | $0.00064 | $0.00920 |
| Sonnet 5 | $0.00026 | $0.00368 |
| Haiku 4.5 | $0.00013 | $0.00184 |
Grade A, and why
ci-code-review scanned grade A with 0 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 2d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 141 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Provide a code review for the given pull request.
Follow these steps:
-
Eligibility Check (Sonnet): Check if the PR (a) is closed, (b) is automated/trivial. If so, stop.
-
PR Analysis (Fable): View the PR and return:
- Summary of the change and its purpose
- List of new functions, types, enums, or abstractions introduced
- For each new abstraction: its name, stated purpose (from comments/docs), and intended usage contract
-
Parallel Deep Review (4 Fable agents):
Pass the PR summary and new abstractions list to each agent.
a. Semantic Consistency Agent: For each new function/type/enum introduced:
- Read its definition, documentation, and any comments describing when/how it should be used
- Find ALL usages of that abstraction within the PR
- Verify each usage matches the documented intent
- Flag misuse: e.g., error-handling functions called for wrong error types, validation functions bypassed, enums used inconsistently
- Pay special attention to: panic/error functions (when should they trigger?), unsafe blocks, security-sensitive operations
b. Deep Bug Analysis Agent: Read the full context of modified files (not just diff lines).
- Understand the data flow and control flow around changes
- Check for logic errors, edge cases, off-by-one errors, resource leaks
- Verify error handling is appropriate for each failure mode
- Check that invariants are maintained across the changes
c. Tech Debt Removal Agent:
- Understand new abstractions that are introduced, new functions/types/enums
- Identify possible future usecases for these things and understand whether abstractions meet future requirements
- See which paradigms of Rust (or other language) development are used and whether they apply here
- Identify possible improvements that would benefit long term maintainability of the code
- Be an enjoyer of abstractions: generics, traits, dyn, enums, etc.
d. Security Reviewer Agent:
- Identify whether changes to the protocol do not break soundness
- Identify possible attack vectors that are introduced with these changes
- Check for input validation gaps at trust boundaries (user input, network data, file I/O, IPC)
- Verify authentication/authorization checks are not bypassed or weakened
- Look for injection risks: SQL, command, path traversal, template injection, deserialization
- Check cryptographic usage: hardcoded secrets, weak algorithms, nonce reuse, timing side-channels
- Verify resource limits: unbounded allocations, missing timeouts, denial-of-service vectors
- Check concurrency: TOCTOU races, lock ordering, shared mutable state without synchronization
-
Validate Issues (MANDATORY — do not skip):
After collecting all issues from the 4 agents, validate every issue scored >= 50.
For each issue from the agents:
- For complex logic/semantic issues, reason through whether the bug is real and exploitable.
- For issues you can verify mechanically (e.g., a failing test), prefer direct verification (run the test).
Score each issue 0-100 AFTER validation:
- 0: False positive, doesn't stand up to scrutiny, or pre-existing issue
- 25: Might be real, but couldn't verify. Stylistic issues without explicit guidance.
- 50: Verified real issue, but minor/nitpick. Not important relative to PR scope.
- 75: Verified real issue that will impact functionality. Insufficient existing approach.
- 100: Confirmed real issue that will happen frequently. Direct evidence confirms it.
-
Post comments to PR: Always post a single review to the PR so the author has confirmation that the review ran, regardless of whether issues were found.
- When there are validated issues with score >= 50, include them in the
commentsarray. - When there are none, post the review with an empty
commentsarray and a body that briefly states no issues were found (1-2 sentences, same tone as the comment guidelines below — concise, senior-engineer voice, no scores/severity labels/ceremony).
- When there are validated issues with score >= 50, include them in the
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.
- 2d ago First seen · 141 lines · 129 tokens per session scan A 677454d1c6b1
ci-code-review is a skill published in the GitHub repository a16z/jolt (1,020 stars, last pushed 2d ago), licensed Apache-2.0. It adds 129 tokens to every session and 1,840 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
linea-dependency-maintenance
Safely plan and execute dependency maintenance for JavaScript/TypeScript (npm, pnpm) and GitHub Actions, including npm lockfiles, pnpm workspaces, catalogs, overrides, SHA-pinned action versions, release-age policies, audits, CI validation, Dependabot boundaries, PRs, and GitHub tracking issues. Use whenever the user…
pr-content-review
Review docs.linea.build content for editorial compliance (voice, terminology, naming, formatting, frontmatter) before opening or merging a pull request. Use before submitting a PR, when reviewing someone else's draft, or when asked to audit existing pages.
starknet-js
Use when writing or debugging JavaScript/TypeScript that interacts with Starknet through the starknet.js SDK — building Call objects or calldata, encoding/decoding Cairo types (felt252, u256, structs, arrays, spans, ByteArray, Option/Result/custom enums), or working with contracts, accounts, providers, transactions…
EMILIA Trust Verification
Verify the authenticity of AI-agent authorization receipts and human-device signoffs. Use this whenever a user shares a "trust receipt", an "authorization receipt", a "signoff", or WebAuthn/passkey approval evidence and asks whether it is valid, genuine, or tampered with. Pairs with the public EMILIA Protocol MCP…
Cryptographic Analysis & Assessment
SSL/TLS auditing, cipher suite analysis, hash algorithm identification, encryption implementation review, and cryptographic weakness detection in code.
author-doc-page
Scaffold or draft a new docs.linea.build page to editorial standards. Use when creating a new page, writing a first draft, or helping a contributor who isn't a professional writer produce content that meets the documentation team's expectations.