Skills are how Octopus tells the agent what to do and how to do it
disciplined. Each skill is a small Markdown file with frontmatter
that triggers engagement based on the user’s intent — a description
that says “use when X”, optional triggers: for auto-engagement on
specific file patterns, and a body that walks the agent through the
right sequence of steps.
Why skills, not just rules
Rules describe what good code looks like (coding style, security,
patterns) — they’re loaded into every conversation as standing
context. Skills describe how to perform a task — they’re loaded
on demand when the user’s intent matches the skill’s description.
The split keeps the active context lean: a rule about Conventional
Commits is always loaded, but the full PR-feedback discipline only
loads when you’re actually receiving feedback on a PR. Skills are
the “engage when needed” layer.
Granularity
Per the docs-site rule, only skills with non-trivial design
decisions get a detail page. Stack-specific patterns, thin
wrappers, and one-trick utilities are listed in the overview tables
without a page of their own.
Audits
Pre-merge audit gates. Each runs read-only and produces a severity-
tiered report (⛔ block / ⚠ warn / ℹ info). The audit-all
composer runs the relevant ones in parallel.
| Skill | Audits | Detail |
|---|
audit-all | Composer for the four audits below — shared file discovery, parallel execution, consolidated report. | → |
audit-security | Hardcoded secrets, injection risks, public route auth gaps, dependency CVEs. | → |
audit-money | Money types, rounding, cents-not-round tests, env-var consistency, payment idempotency, webhook signature verification, fee-disclosure coupling. | → |
audit-tenant | Multi-tenant data-scope enforcement — query filters, raw SQL, new entity configs, controller ownership, admin endpoints. | → |
audit-config | Agent-configuration freshness — model-specific assumptions, stale dates, phantom skills, deprecated paths. | → |
audit-grounding | Signal-only divergence from the source of truth — invented conventions, unsupported domain facts the formatter can’t see. | → |
audit-fleet | Cross-repo adoption + drift — each repo’s config vs its declared target (baseline + stack profile + adoption tier) from fleet.yml. Signal-only; feeds fleet-bootstrap. | → |
audit-contracts | Frontend/backend contract drift — endpoints, DTOs, enums, status codes, auth rules, params. | → |
Implementation workflows
| Skill | Engages when | Detail |
|---|
implement | Any code-editing task — default on every implementation. | → |
debug | The user reports a bug, test failure, or unexpected behaviour. | → |
prototype | The user wants to flush out a design before committing — throwaway code in service of decisions. | → |
test-tdd | Implementing a new feature with explicit red/green/refactor loop. | — |
test-e2e | Adding an end-to-end test that hits real services. | — |
test-component | Adding a component-level test (RTL) — behaviour over implementation. | — |
backend-patterns | Backend implementation work — stack-aware layering. | — |
frontend-patterns | Frontend implementation work — component design, state, accessibility. | — |
Refactoring
| Skill | Engages when | Detail |
|---|
refactor-deepen | The codebase has shallow modules that should be consolidated. | → |
Doc lifecycle
The full RFC → Spec → ADR → Knowledge chain. Each step is a skill
the corresponding command invokes.
| Skill | Output | Detail |
|---|
doc-lifecycle | Routes between RFC / Spec / ADR / Knowledge based on scope and risk — the entry-point skill for documentation work. | → |
doc-adr | A new Architecture Decision Record in docs/adr/. | → |
doc-plan | A bite-sized TDD-style implementation plan from an approved spec. | → |
doc-subcontext | Per-module CLAUDE.md for monorepos — inherits from parent. | → |
doc-api | Integrator-facing API reference + OpenAPI kept faithful to the code — validate the contract or regenerate the docs. | → |
doc-align | Reconcile drift between code and docs. | — |
doc-prd | PRD published to the project issue tracker. | — |
Knowledge & planning
| Skill | Engages when | Detail |
|---|
continuous-learning | Capturing insights worth keeping across sessions; team mode aggregates recurring review feedback fleet-wide into rule candidates. | → |
standards | Self-serve “what’s our standard for X, and why” — answered from ADRs, rules, CONTEXT.md, knowledge; routes to authoring on a miss. | — |
definition-of-done | The team’s Definition of Done as a versioned artifact — create/update it, or validate a diff against it. Wired into codereview, signal-only. | → |
onboarding | Ramp a new engineer onto a repo’s standards, architecture, and workflow — a scoped, ephemeral ramp that presents the map-system deck. | → |
plan-backlog | Auditing the roadmap and plans for staleness, orphans, broken links. | → |
context-budget | The conversation is heavy on tokens and needs pruning. | — |
context-handoff | Handing the current session to another agent. | — |
map-system | Mapping a repo — a quick textual orientation, or a themed self-contained HTML deck of the whole repo (the default). | → |
fleet-bootstrap | Converge 6+ repos onto a layered standard (baseline + stack profile + adoption tier) from one fleet.yml — multi-stack and legacy first-class; dry-run, non-destructive. | → |
interview | Stress-testing a plan via grilling. | — |
council | Pressure-testing one high-stakes decision from five lenses with anonymous peer review and a chairman verdict. | → |
triage-issues | Routing incoming issues through a state machine. | — |
Launches
| Skill | Output | Detail |
|---|
launch-feature | Multi-channel launch kit from a completed feature (RM / spec / PR). | → |
launch-release | Themed release announcement kit for existing users — landing page, email, Slack, Discord, banner, X thread, deck. | → |
| Skill | Engages when | Detail |
|---|
scaffold-skill | Creating a new skill end-to-end. | Covered in /commands/scaffold-skill/ |
compress-skill | Shrinking a SKILL.md without changing semantics. | — |
delegate | Routing a task to a specific role. | — |
Stack patterns
| Skill | What it adds |
|---|
dotnet | .NET / C# stack-specific guidance — async/await, EF Core, naming, file structure. |
Receiving code review
| Skill | Engages when | Detail |
|---|
respond-to-review | Receiving feedback on a PR — the verify / ask-for-evidence / disagree-with-reasoning triage. | Covered in /commands/pr-comments/ |
Source: skills/
