Skip to content

cipherstash/company-skills

v0.11.0MIT AND Apache-2.0 AND OGL-UK-3.0

Shared company agent skills for Codex and Claude Code.

authoring-pull-requests

Takes finished work from the working tree to a draft pull request on a CipherStash GitHub repository — a type-prefixed branch, signed conventional commits that explain why, the repository's own pre-PR checklist and lint/build/test commands, and a plain-language PR body (Summary, Changes, Verification, Related, Review notes) with honest verification. Use when committing work, opening or updating a pull request, writing a PR title or body, re-signing or rebasing a branch that has an open PR, or asked to "ship", "push this up", or "raise a PR". Opens PRs as drafts and never merges, closes, or marks one ready without being asked. Not for fixing review findings or auditing review comments on an existing PR (review-fix, review-audit) or for filing GitHub issues (filing-github-issues).

filing-github-issues

Drafts and files GitHub issues on CipherStash repositories — searches open and closed issues in the current repo and its siblings first, verifies every mechanical claim with file:line evidence, picks the right repo (upstream cause versus consumer impact), and writes a Background / Problem / Proposal body in plain wording a reader new to CipherStash can follow. Use when drafting or filing a GitHub issue on a CipherStash repository, or when asked to write up a bug, gap, or proposal as a GitHub issue. Files only when the user asked to file or approves the draft. Not for Linear issues or tickets, and not for pull requests (authoring-pull-requests).

plain-language

Writes, rewrites, and reviews prose a person reads in plain language, following ISO 24495-1:2023, ISO 24495-3:2026, and the GOV.UK content guidance, and checks the result with a bundled checker. Use when writing a document, README, plan, spec, report, findings, or summary for people to read; when rewriting or reviewing text that is hard to read or full of jargon, metaphor, idiom, or cliché; and when the user mentions plain language, plain English, or ISO 24495. Not for writing or changing code or configuration, or for generated reference output such as API docs, though code comments and commit messages are in scope, and a repository's own style guide wins where it is stricter or where it sets a different house style.

review-audit

Collate a pull request's existing review feedback — inline threads, review bodies, and conversation comments from people and bots — verify each finding against the PR's head commit, and report which remain unresolved and which are addressed and can be resolved. Use when asked what review feedback is still outstanding on a PR, which review threads can be resolved, or to audit PR comments against the code. Takes a PR number or URL. Does not fix findings (review-fix), run a new review, or cross-check new findings against the PR (review-cross-check).

review-cross-check

Verify new code review findings — this session's review or ones pasted from parallel agents — against a pull request's head, cross-check the confirmed ones against the PR's existing review feedback, and report which findings are new, which materially add to existing ones, and which are already covered or refuted. Use after reviewing a PR, or when asked to check findings against existing PR comments or drop findings reviewers already raised. Takes a PR number or URL. Does not perform the review, fix code, or post to the PR (review-post).

review-fix

Verify and fix code review findings — this session's review plus any pasted from parallel agents — each confirmed behavioural defect fixed red first with a regression test wherever one can be written. Use when asked to fix, or work through, the findings of a code review run in this session, a PR's unresolved review feedback as audited by review-audit, or findings pasted from review agents. Does not perform the review, reply to or resolve PR review threads, or audit a PR's existing feedback (review-audit).

review-post

Posts a finished code review to a GitHub pull request as a single review — one inline comment per finding, tagged with its source, plus an attestation summary naming the commit reviewed, the host, models, and skills, with blocking issues first — posted even when there are no findings. A finding that adds to an existing review thread is posted as a reply on that thread. Use when asked to post, publish, or leave review findings or a review attestation on a PR. Takes a PR number or URL. Does not perform the review, verify findings, cross-check them against existing comments (review-cross-check), or resolve threads (review-audit).

run-report

Keeps a log of every problem an unattended run hits — errors, failed commands, retries, fallbacks, workarounds, skipped or partial steps, missing skills or access, and assumptions made in place of instructions — then publishes a run report as a Claude artifact, or in the final message where artifacts are unavailable, with the run's status, an analysis of each problem, and its impact on what the run produced; a clean run still publishes one. Use when a scheduled task's or routine's prompt lists this skill or asks for a run report, or when asked to report what went wrong in a run. Does not diagnose a single failure, post review findings (review-post), or file issues (filing-github-issues).

rust-sdk-dx

Applies CipherStash conventions for public Rust APIs in libraries, SDKs, and shared crates — thiserror error enums, typestate builders, required derives, newtype identifiers, ergonomic signatures, rustdoc sections, tokio async, crate layout, and feature flags. Use when writing or changing Rust library code that other developers consume, such as a crate with lib.rs, a public type or function, an error type, a builder, or module structure. Does not apply to application binaries, internal scripts, or non-Rust files, and defers secret handling to rust-security.

rust-security

Applies CipherStash company rules for Rust code that handles secrets, cryptography, and authentication — the crate-level security lint block, opaque secret types with no AsRef or Deref, crypto library order (vitaminc, then aws-lc-rs, then RustCrypto), one encryption context per purpose, udigest for structured hashes, opaque errors at crypto and HTTP boundaries, JWT verification, constant-time encoding, and logging discipline. Defers vitaminc types and APIs (Protected, Equatable, AEAD, PRF, SafeRand) to the vitaminc skill or docs.rs. Use when writing or changing Rust code that encrypts or decrypts, derives or stores keys, defines a token, key, password, or other secret type, compares or hashes secret values, generates security-relevant randomness, verifies credentials or JWTs, or serializes or logs data that may contain secrets. Does not apply to Rust code with no secret or crypto handling, non-Rust code, or general public API design, which rust-sdk-dx owns.

rust-testing

Applies CipherStash conventions for Rust tests — running unit and integration tests with nextest and its test groups, nested given_* scenario modules with expectation-style names, assertion messages, randomised test data with fake and Dummy impls behind a test-utils feature, per-test mocktail HTTP mocks, paused-time tokio tests, concurrency stress tests, and testing napi-rs boundaries. Use when writing or changing Rust tests (#[test],

vanta-remediate

Patches the dependency vulnerabilities tracked by Vanta Linear issues — for each "[Vanta] Remediate {package} to {version} ({repo})" issue, finds every lockfile location that resolves a vulnerable version, bumps it with the right package manager, verifies the fix landed, opens a GitHub PR, and moves the Linear issue to In Review. Also assesses "[Vanta] Assess" container OS-package issues on request. Use when asked to fix, patch, remediate, or open PRs for Vanta or Dependabot vulnerability findings, work a CIP Remediate issue from the AppSec project, or run the scheduled Vanta remediation, including unattended routine runs. Does not query Vanta or file issues (vanta-triage does that), review existing PRs, fix AWS configuration findings, or perform feature upgrades.

vanta-triage

Reconciles open Vanta vulnerability findings (Dependabot package alerts and AWS Inspector container findings) with the Linear AppSec tracking issues — filing, retargeting, reopening, and closing issues so Linear matches Vanta. Use when asked to check Vanta, look for new security alerts or vulnerabilities, sync Vanta with Linear, file Vanta issues, or run the scheduled Vanta triage, including unattended routine runs. Does not patch repositories or open PRs (vanta-remediate does that), and does not handle Vanta AWS configuration tests.