laimis91/assistant-framework
Assistant Framework skills: workflow, review, verification, TDD, debugging, research, security and more, for Claude Code and Codex.
Decides what to do with an incoming security advisory, advisory draft, or vulnerability report such as a GHSA or CVE report: close, keep open, or keep open but narrow, checked against shipped releases. Use when triaging or dispositioning a reported vulnerability; not for threat modelling, code or dependency vulnerability audits (assistant-security), implementing a known fix, or writing release notes.
Audits CI and test suites for false passes, discovery gaps, stale fixtures, brittle or duplicate tests, and cost. Use when a suite is weak, slow, or hard to trust; not for fixing one failing test or CI run.
Clarify ambiguous or multi-intent requests. Use when the user asks to untangle, structure, or restate what they mean.
Diagnose an unknown failure by reproducing and isolating before fixing. Use for debugging, root causes, flaky tests, or unexplained breakage.
Create or update README, API, architecture, changelog, or migration docs, or diagrams of existing code. Use for explicit documentation work or a diagram of existing code.
Map an unfamiliar codebase and its patterns. Use for onboarding, project orientation, or explicit codebase-learning requests.
Gathers and synthesizes source-backed information. Use for research, topic investigation, current evidence, or requests to compare options.
Reviews code and diffs (the current uncommitted changes, named files, a branch or a pull request), specs, plans, tests, documents, Jira tickets, and refinement material, fixes actionable findings, and runs one fresh re-review, or reports only in audit mode. Use for code review, C# pull requests, readiness checks on such material, and the workflow Review phase; QA runs only when required. Assessing a skill itself belongs to assistant-skill-creator; formal review of a delivered diff stays here.
Analyzes threats, vulnerabilities, dependencies, and attack surfaces. Use for security work, and before changing auth, external input, secrets, sensitive data, persistence, network or shell calls, dependencies or config, or external integrations.
Creates, updates, assesses, or reviews skills (V1 or a plain SKILL.md), or checks registered upstream guidance read-only. Use for skill contracts, compliance, or authoring, and before any edit to a SKILL.md, activation case, or skill eval. Delivering a skill change in a repository also needs assistant-workflow and assistant-verification.
Answers one observable feasibility question by running a small disposable experiment outside tracked files, then reports a validated, partial, or invalidated verdict and builds no production change. Use when the user asks to spike, try out, or prove whether an approach can work and only execution can settle it, not reading sources, scoring trade-offs, building the real change, or diagnosing a failure.
Use for tests-first TDD requests, project TDD rules, or workflow-selected TDD checks.
Reason through architecture decisions and complex trade-offs, and generate and rank options. Use for explicit stress tests, debate, first principles, hypotheses, quick improvement scans, or explicit brainstorm requests.
Explores a running web or app UI for unknown defects through the session's browser tools: flows, valid, invalid and empty states, console and network errors, screenshots, and reproducible issue reports with severity and coverage. Use when the user asks to test, dogfood, or click through a running app or page in a test environment. Not for writing UI tests, diagnosing a known failure, code review, or acceptance verdicts.
Analyzes and selects proportionate testing, validation, or evaluation for a change, or assesses whether existing evidence is sufficient across code, skills, plugins, and non-code artifacts. Use on request, and before the first edit of any non-trivial feature, fix, refactor, migration, or skill change, including after plan approval, review feedback, CI failures, or resuming, even when assistant-workflow or assistant-debugging has already loaded. Not for running an existing test command or tests-first requests (assistant-tdd).
Runs repository changes end to end: prepare, plan, build, or resume repository task state for feature/epic/story technical preparation (including Architecture Packs), implementation, fixes, migrations, refactors, artifacts, backlog items, and intended-change ideas. Use before the first edit, commit, or PR on any turn that changes code, tests, specs, or skill files, even a terse one: plan approval, post-compaction continuation, review comments or findings, CI fixes, unprocessed work ready for a PR. Pairs with assistant-verification before editing.