Menu
Pick the skill or flow that fits your situation. A router over the skills in this repo.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Recurring maintenance pass over automated dependency-bump PRs (Dependabot, Renovate, or similar) โ rebase, relock, verify, and report what's ready to merge. Meant to be handed to a time-based loop or schedule, not run once.
Post-deploy verification loop โ poll a rollout until every instance is on the new version and healthy, then run one smoke check against a real user path. Meant to be handed to a time-based loop with a timeout, not polled by hand.
Reference for designing agent loops โ cycles of work that repeat until a stop condition is met. Covers the four loop shapes, writing completion criteria, carrying state and isolating work across cycles, and what running unattended still leaves on you.
Recurring check-in on a stack of dependent PRs โ rebase children onto updated parents, surface CI state, flag anything waiting on a human-only gate, and report only what changed since last time. Meant to be handed to a time-based loop, not run once.
Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes โ Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/PRD asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".
Shared vocabulary and principles for designing deep modules. Use when the user wants to design or refine a module's interface, hunt for deepening opportunities, decide where a seam belongs, make code more testable or AI-navigable, or when another skill needs the deep-module vocabulary.
| name | which-skill |
| description | Pick the skill or flow that fits your situation. A router over the skills in this repo. |
| disable-model-invocation | true |
You won't hold every skill in your head, so route through this instead of guessing.
A flow is a path traced across the skills. Most work follows a single main flow, with two on-ramps feeding into it. The rest are either standalone tools or a vocabulary layer that operates beneath everything else.
The route most work takes. You have an idea; you want it built.
/grill-with-docs โ sharpen the idea through interview. Start here when there is a codebase: it keeps state, folding what it learns back into CONTEXT.md and ADRs. (No codebase yet? Use /grill-me โ see Standalone. Both drive the same /grilling primitive; only grill-with-docs leaves a written trail.)
Branch โ can every open question be closed in conversation? If any question demands a runnable answer (state, business logic, a UI you have to lay eyes on), take a detour through a prototype, with /handoff carrying context in each direction (see Crossing sessions):
/handoff out, then start a clean session against that file,/prototype to settle the question with disposable code,/handoff the findings back, and cite them from the original idea thread.Branch โ is this a multi-session build?
/to-prd (fold the thread into a PRD) โ /to-issues (cut the PRD into issues anyone can grab independently). Since the issues stand alone, reset context between them: open a fresh session per issue and start /implement with the PRD plus the one issue to work on./implement right here, without leaving this context window.Either path, /implement delivers each issue by driving /tdd underneath โ one red-green slice at a time โ then finishes with /code-review, a two-axis pass (Standards + Spec) over the diff, before it commits. Pull in /tdd by itself when you simply want a concrete behaviour built test-first without a full spec, and /code-review by itself whenever you want a branch or PR checked against a fixed baseline.
Hold steps 1โ3 inside one continuous context window โ don't compact or clear until /to-issues is done โ so the grilling, the PRD, and the issues all rest on the same line of thinking. Each /implement then begins clean, taking the issue as its starting point.
What bounds this is the smart zone: the window (roughly 120k tokens on current frontier models) inside which the model still reasons crisply. If a session nears that edge before /to-issues, don't grind on through degraded reasoning โ /handoff and pick up in a fresh thread.
A starting situation that generates work, then merges onto the main flow.
Bugs and requests stacking up โ /triage. It walks issues through their triage roles and turns out agent-ready issues, which /implement later collects.
Triage is for issues you didn't author โ bug reports, incoming feature requests, whatever lands raw. Issues that /to-issues produced are already agent-ready, so skip triage for those.
Something's broken โ /diagnosing-bugs. Built for the stubborn ones: the bug that shrugs off a first look, the intermittent flake, the regression that slipped in between two known-good points. It won't theorise until it holds a tight feedback loop โ one command that already goes red on this bug โ then lands the fix behind a regression test. Its post-mortem hands off to /improve-codebase-architecture when the real lesson is that no clean seam exists to pin the bug down.
Recurring work, or the main flow with nobody watching in real time.
/designing-loops โ reference for the four loop shapes and how to write a completion criterion that actually stops the loop. Read this before wiring up any of the below./dependency-update-loop, /stacked-pr-loop, /deploy-verify-loop โ ready-to-hand-to-a-schedule recipes for the recurring maintenance that piles up around shipped code: bot PRs, PR stacks, and the window right after a deploy.Not feature work โ upkeep.
/improve-codebase-architecture โ reach for it whenever you have a spare moment to keep the codebase pleasant for agents to work in. It surfaces deepening opportunities; choosing one spawns an idea you can carry into the main flow at /grill-with-docs. It's the survey that turns up candidates; /codebase-design (below) is the bench where you shape the one you picked.Turn a spec or a settled design into the checks that keep an implementation honest. Each distils the current thread into an in-repo doc; run them once the design is firm, before or alongside /implement. They compound โ invariants become fuzz oracles, and both sharpen the conformance suite.
/to-invariants โ the properties that must always hold. Splits into design-level system invariants and enforceable ones, each mapped to a check or flagged as an unguarded GAP. Start here; the other two lean on its output./to-fuzz โ the targets worth fuzzing, each pinned to an oracle (differential, round-trip, invariant, crash). Reuses /to-invariants output directly โ every enforceable invariant is a ready-made oracle./to-conformance โ a spec's normative requirements (MUST/SHOULD/MAY) turned into implementation-independent tests any conforming build must pass, with a traceability matrix and flagged spec ambiguities.Two model-invoked references that sit below the other skills โ each the single source of truth for its vocabulary. Call them directly when the words, not the process, are the sticking point; or let the skills above draw them in.
/domain-modeling โ tighten the project's domain language: challenge a fuzzy term, untangle an overloaded word ("account" pulling triple duty), pin a hard-to-reverse decision into an ADR. It's the live discipline /grill-with-docs drives to keep CONTEXT.md a clean glossary./codebase-design โ the deep-module vocabulary (module, interface, depth, seam, adapter, leverage, locality) for shaping a module's form: plenty of behaviour behind a small interface at a clean seam. Both /tdd and /improve-codebase-architecture speak it./handoff โ when a thread is full or you need to peel off (say, into a /prototype session), this compresses the conversation into a markdown file. You don't resume in place โ you open a new session and point it at that file to ferry the context across. It's the bridge between context windows, working either way. Use it when you want a fresh session but need the current conversation kept intact./compact (built-in) โ stay in the same conversation and let the earlier turns be summarized. Use it at deliberate breaks between phases, when losing the verbatim history is fine. Don't compact mid-phase โ the agent can lose the thread. /handoff forks; /compact continues.Off the main flow entirely.
/grill-me โ the same unrelenting interview as /grill-with-docs, but for when there's no codebase. Stateless: it writes nothing locally, builds no CONTEXT.md. Reach for it to sharpen any plan or design that doesn't live in a repo./prototype โ a small, disposable program that answers one design question: does this state model feel right, or what should this UI look like. Throwaway from the first line โ keep the answer, delete the code. It's the detour in step 2 of the main flow, but reach for it any time a design question is hard to settle on paper./research โ hand the reading legwork to a background agent: it chases a question through primary sources, then drops a cited Markdown file into the repo. Keep working while it reads. What it produces is something to bring into the main flow at /grill-with-docs โ research feeds the thinking, it doesn't stand in for it./teach โ learn a concept across several sessions, treating the current directory as a stateful workspace./writing-great-skills โ reference for writing and editing skills well./setup-skills โ run it before your first engineering flow to configure the issue tracker, triage labels, and doc layout the other skills expect. Custom issue trackers work too.