--- name: using-skills description: Routes the current work to the right Keystone skills and chains them in order. Use at session start or when unsure which process applies. Triggers - which skill should I use, where do I start with this task, what is the process here, hangi skill, nereden başlayayım. --- # Using Skills Keystone is a process library. Each skill encodes how one kind of engineering work is done properly. This meta-skill decides which one applies and in what order — picking wrong is worse than picking none, because a half-matching process produces confident, wrong-shaped work. ## Read first, every session 1. `CLAUDE.md` — the invariants. They outrank every skill. 2. The project's `CONSTRAINTS.md` — the numbers. If absent, run `/constraints`. 3. Existing ADRs (`docs/adr/`) — decisions already made are not re-litigated mid-task. ## Router | Situation | Skill | |---|---| | Request is vague, or you are about to guess what someone wants | `requirements-interrogation` | | Intent is clear, nothing is written down | `spec-driven-development` | | Spec exists, work needs sequencing | `planning-and-task-breakdown` | | A choice will be expensive to reverse | `architecture-decisions` | | The words in the code don't match the words in the business | `domain-modeling` | | Modules are tangled, a change touches everything | `boundaries-and-dependencies` | | Designing anything another system calls | `api-and-contract-design` | | Choosing or shaping storage, indexes, schema | `data-modeling-and-storage` | | Network calls, queues, partial failure, retries | `distributed-systems-and-resilience` | | Shared mutable state, locks, transactions, races | `concurrency-and-consistency` | | Building user interface | `frontend-architecture` | | Writing the code for a planned task | `incremental-delivery` | | Any logic or bug fix | `test-driven-development` | | Deciding what to test and at which level | `testing-strategy` | | Code works but is hard to read or change | `refactoring-and-simplification` | | Commits, branches, history, PR hygiene | `git-workflow` | | Before merge | `code-review` | | Something broke or behaves unexpectedly | `debugging-and-root-cause` | | Something is slow, or a budget is at risk | `performance-engineering` | | Untrusted input, auth, secrets, sensitive data | `security-and-threat-modeling` | | You cannot tell what production is doing | `observability-and-slos` | | Cutting a version, shipping, rolling back | `release-engineering` | | Removing or replacing something in use | `deprecation-and-migration` | | Production is on fire, or just stopped being | `incident-response` | | Adding a library, model, dataset or copied snippet | `dependency-and-license-policy` | | A config value, API key, `.env` file or leaked secret | `configuration-and-secrets` | | Adding or trusting an MCP server, plugin, hook or agent permission | `agent-and-toolchain-security` | | Writing docs a future person will rely on | `documentation-and-knowledge` | | Putting a model in the product | `ai-integration-guardrails` | | Setting or enforcing the quality bar | `constraint-driven-development` | | Configuring what the agent knows about this repo | `context-engineering` | | Several agents on one task, parallel work, or a loop that will not stop | `agent-orchestration` | | Adding to Keystone itself | `skill-authoring` | ## Chains Chains matter more than individual skills — most real failures happen between stages, not inside them. **New capability** `requirements-interrogation` → `spec-driven-development` → `architecture-decisions` (only if a choice is costly to reverse) → `planning-and-task-breakdown` → per task: `incremental-delivery` + `test-driven-development` → `code-review` → `release-engineering` → `observability-and-slos` **Integrating an external system** `api-and-contract-design` (their contract, your adapter) → `distributed-systems-and-resilience` (timeouts, retries, idempotency) → `testing-strategy` (contract tests, recorded fixtures) → `observability-and-slos` **"It's slow"** `performance-engineering` (measure, find the real bottleneck) → then whichever of `data-modeling-and-storage`, `concurrency-and-consistency`, `frontend-architecture` the measurement points at — never the reverse order **Inherited code you must change** `boundaries-and-dependencies` (map what depends on what) → `testing-strategy` (characterization tests as a safety net) → `refactoring-and-simplification` → then the actual change **Replacing something live** `deprecation-and-migration` → `incremental-delivery` (expand/contract) → `observability-and-slos` (shadow comparison) → `release-engineering` ## Rules - Multiple skills apply more often than one. Apply all of them; do not pick a favorite and drop the rest. - Reading a skill costs little; applying the wrong process costs a rewrite. - No skill fits? Fall back to `CLAUDE.md` invariants and say so. A gap felt three times is a skill worth writing (`skill-authoring`). - A skill never overrides `CLAUDE.md`, `CONSTRAINTS.md`, or an accepted ADR. ## Exit checklist - [ ] `CLAUDE.md` and the project's `CONSTRAINTS.md` were read this session - [ ] Every skill that applies to the current work has been identified, not just one - [ ] The chain, not only the individual step, has been considered - [ ] Any gap where no skill fits has been stated rather than improvised over