Claude Skills by luk-s12

Data modeling and safe-migrations guide, agnostic of engine (relational and non-relational). worktree-agent-data MUST apply it when designing/implementing, classifying each finding by severity and citing concrete evidence before declaring it blocking - same criterion as security-guide, so that a schema change doesn't break already-deployed code or existing data.

Launches the research agent (spike mode) on a specific question or topic - a library, a technical approach, or some base-repo behavior - without implementing anything. Publishes the finding as an Announcement in .orchestrator/projects/<project>/announcements/ so future User Stories can take advantage of it.

Migrates a pre-multiproject Legion installation after Git has updated the checkout. It resumes the singleton-to-multiproject bootstrap and never updates Git or installed modules.

Orchestrates project-scoped User Stories with one worktree per story, simple cross-session claims, design approval, implementation and adversarial review.

Lifecycle management for an already-installed module — uninstall (with confirmation, deferred if active stories still reference it), activate (undo a pending deprecation), and renegotiate (reopen a provides_rules verdict for the current project). Use /new-module for installing a new module instead.

Assistant for recording in .orchestrator/projects/<project>/lessons-learned.md an incident (in production or any other environment) caused by a business rule that hadn't been accounted for. Asks for the minimum data, fills in the date on its own, formats it, and if the responsible User Story is identifiable updates .orchestrator/projects/<project>/reputation.md - so future User Stories touching the same code zone find it before repeating the mistake.

Installs an external module (a Claude Code project — an agent, optionally with skills — that plugs into an orchestration as a story gate, a standalone generator, or a story implementer that writes code directly). Clones the repo, validates its manifest, runs a best-effort risk scan (tools, network patterns, vulnerable dependencies), shows the user a preview, and only registers it in modules/registry.md once accepted.

Assistant for splitting a high-level objective (not yet shaped into User Stories) into several concrete User Stories. Researches the base repo, proposes a split with evidence, confirms it with the user, persists the decision in .orchestrator/projects/<project>/objectives/, and specifies each resulting User Story with the same depth as /new-story - one at a time, saving them into requirements/<project>.md as soon as each one is confirmed.

Assistant for creating a User Story in requirements/<project>.md. Asks for the ID and description, analyzes the description against the base repo's real code, challenges ambiguities and possible bugs with questions to the user, and only then writes the refined User Story to the file — through a semantic preview checkpoint and a final-text checkpoint, keeping a persistent working record per story.

Design quality and code-smell detection guide, agnostic of language and paradigm (object-oriented, functional, imperative/procedural; works for frontend and backend). Agents in every worktree MUST apply it before implementing and before reporting FINALIZED, so that no User Story produces ugly or divergent code. The orchestrator uses it as a tiebreaker in architectural decisions.

Invokes a type generator module on demand — not story-scoped, runs outside the /legion cycle entirely. Reads the base repo (or a specific story's worktree) and produces a standalone artifact in modules/output/. Use /new-module first to install the module.

Security audit guide, agnostic of language and framework, based on the OWASP Top 10. worktree-agent-security MUST apply it when auditing, classifying each finding by severity and citing concrete evidence before declaring it blocking - so that the lack of context of a narrow audit doesn't stall a User Story for no real reason.