Skills Catalogue
Auto-generated by
packages/core/skills-catalogue.tsfrom eachSKILL.mdfrontmatter. Do not hand-edit; runnpm run catalogafter adding or recategorising a skill.
Skills are on-demand runbooks. The 2 skills without disable-model-invocation: true keep
their one-line description in context; the other 36 are surfaced on demand by
skill_search and trigger hints, and any SKILL.md body loads when the task matches
(or via /skill:<name>). See
Writing skills for the format and rationale.
Coding & general workflow
- codebase-navigation — Orient in a repository and find/read code with the fewest tool calls.
- coding-agent-workflow — Reliable small-step coding — repo maintenance, bug fixes, features, refactors.
- general-agent-workflow — A reliable inspect-plan-act-verify loop for general (non-coding, non-engagement) tasks.
- patch-hygiene — Keep edits scoped, clean, and reviewable.
- supply-chain-review — Review dependency, image, and MCP server supply-chain risk.
- task-decomposition — Break work into small, reversible, independently-verifiable steps.
- verification-loop — Choose and run the smallest check that can disprove a change, then fix-and-rerun.
Efficiency & anti-loop
- self-reflection-and-recovery — Spend few tool calls, notice when you are looping or not progressing, and break out by reflecting, changing approach or delegating.
Context & small models
- context-management — Keep long work recoverable across compaction, resets and session boundaries — maintain a checkpoint, compress without losing the thread, and pick the right control (/compact, /fork, /clone, /new).
- small-model-execution — Execute reliably on a small/local model with limited context.
Memory
- memory — Save, recall, correct and forget long-term memories in the vault.
Orchestration & recovery
- agent-orchestration — Delegate non-trivial work to scout/planner/implementer/reviewer subagents with exactly the context each needs, or run a saved workflow.
- dynamic-agent-synthesis — Assemble a deterministic least-privilege Conductor specialist from catalogued skills and policy inputs.
- engagement-conductor — Drive durable Conductor engagement lifecycles across sessions.
- recovery-debugging — Recover after a failed command, bad patch, or interrupted session.
- workflow-authoring — Write and run a JavaScript script workflow for the
workflowtool (hooks, pipeline-by-default, adversarial verify, judge panel, loop-until-dry, completeness critic, resume) or run a kit built-in (understand, review-changes, design-panel, research-sweep, migrate).
MCP & governance operations
- action-card-builder — Build narrow, hash-bound engagement action cards for human approval.
- mcp-only-operations — Keep all tool use behind MCP and local governance in an engagement/governed context.
- mcp-tool-use — Use MCP tools safely and efficiently through the governed adapter.
- scope-roe-governance — Enforce deterministic scope and rules-of-engagement checks before any engagement action.
- tool-policy-classifier — Classify MCP tools into local approval and risk classes.
Engagement assessment
- api-testing — Plan and review authorized API security tests.
- authz-idor-testing — Plan and run low-impact authorization / IDOR validation on an authorized engagement.
- burp-web-triage — Triage Burp-derived web evidence into endpoint observations and hypotheses.
- code-security-review — Review source code for security issues on an authorized, source-assisted assessment.
- endpoint-inventory — Build an endpoint inventory from normalized evidence during an API/web assessment.
- hypothesis-lifecycle — Track assessment observations from candidate hypothesis to finding-ready.
- methodology-mapper — Map observations and hypotheses to methodology coverage (OWASP/WSTG/PTES).
Evidence & reporting
- daily-debrief — Run a conversational end-of-day assessment debrief — assess your own notes, interview the tester to fill the gaps, then reconcile findings and organise notes into the house report format.
- evidence-review — Review evidence quality, hashes, and finding readiness.
- finding-report — Render a validated finding into the house report format — data block, risk statement, background, overview, remediation, evidence.
- finding-writing — Write evidence-backed, professional engagement findings.
- independent-finding-validation — Validate specialist findings or claimed-done results through Conductor's causally independent validator path.
- report-export-review — Check finding and report readiness before export.
Documentation
- documentation-workflow — Produce operator docs that stay aligned with real runtime behavior.
Kit maintainers
- governance-extension-maintainer — Maintain the Pi engagement-governance extension and controlled evidence/approval tools.
- validation-test-maintainer — Keep Luna's validation checks aligned with required profile and runtime contracts.
- wrapper-runtime-maintainer — Maintain the Luna container wrapper without forking upstream Pi core.
Lite profile subset
The lite profile enables 10 of 38 skills, for small local models:
codebase-navigation, coding-agent-workflow, context-management, mcp-tool-use, patch-hygiene, recovery-debugging, self-reflection-and-recovery, small-model-execution, task-decomposition, verification-loop.