Skill catalog

Kiyo has exactly eight public Skills. Each is selected through your host (see Host commands and selectors) and then given a task in plain language. This table is condensed from the framework’s Skill guide.

Skill	Use it to	Input	Expected output	Write / execution boundary
Init	Onboard, analyze, create Memory, inspect readiness	Root or module, explicit onboarding intent, optional preferences	Evidence-backed summary, unknowns, local changes or proposals, activation readiness	Preview reads only; initialize may write authorized Memory, config, and managed bootstrap; never source, tests, dependencies, or global settings
Requirement	Turn a raw request or issue into an engineering requirement	Request, document, or proposed change plus authorized context	ID, objective, behavior, scope, rules and AC, validation, dependencies, security and data, decisions, evidence	Chat default; only a requested specification path may be written
Implement	Feature, bug fix, explicitly scoped refactor	Change intent, behavior and AC, boundaries	Scoped diff, actual verification, review, Memory impact, remaining work	Authorized code, tests, and docs changes plus preflighted checks; no implicit commit, push, PR, deploy, or production database
Review	Assess human or AI changes	Staged or unstaged diff, files, commit range, or accessible PR	Findings with severity, confidence, location, impact, fix idea, and verification idea	Read-only, chat default; no writes, installs, or build/test execution
Test	Find gaps, run checks, or author tests	Requirement, change, or tests plus selected effects	Plan and gaps, actual results, or a test diff, with limits	assess, run, and write have distinct boundaries
Security	Bounded application, Skill, policy, or resource assessment	Supplied target and permitted inspection scope	Evidence, impact and confidence, mitigation, owner, coverage gaps	Read-only; no global scans, credentials, external probes, payload execution, or auto-fix
Architecture	Design question, change impact, or drift	Module or boundary, proposed change, approved decision	Observed structure, approved intent, proposals, deployment unknowns, limits	Read-only; no refactor, migration, or decision/Memory updates
Memory	Show, check, sync, or repair selected context	Canonical store, entries, scope, requested mode	Bounded summary, factual delta, conflict or repair report	show/check write nothing; sync/repair touch only authorized affected entries

Read-only is the default for analysis

Read-only permits necessary authorized inspection, not arbitrary commands. A separately requested report file is its own write scope and never authorizes editing the assessed code or Memory.

Modes are words, not flags

Modes such as Init preview/initialize, Test assess/run/write, Security application/skills/governance/self-check, and Memory show/check/sync/repair are plain intent in your request. No host parses them as arguments.

Skill	Mode	Example request (illustrative)
Init	preview	“Preview onboarding only.”
Requirement	drafting	“Draft requirements for Excel export using repository evidence. Identify unresolved fields and permissions; do not implement.”
Implement	bug fix	“Fix the supplied boundary error and add its regression test; preserve unrelated edits.”
Review	change review	“Review my unstaged changes; do not edit or run project scripts.”
Test	assess	“Assess authorization test gaps in this module; no writes or execution.”
Test	run	“Run the existing approved unit suite after inspecting its script and environment.”
Test	write	“Add tests for these agreed cases in the existing framework.”
Security	skills	“Review this supplied Skill package against AST01–AST10 without executing it.”
Architecture	drift	“Compare ADR-007 with this module; report deviations only.”
Memory	check	“Compare these observations and decisions with this branch.”

What every Skill shares

All eight Skills read the same Core and bootstrap, route through the Workflow Router, and close under the shared Definition of Done. Tiny tasks stay compact; mandatory safety, approval, and evidence checks remain.

Behavioral test catalog

Each Skill has four developer-only behavioral cases (BEH-<SKILL>-01 to 04), plus 16 cross-cutting cases, in tests/behavioral/evaluation/catalog.json. All 48 cases are NOT_RUN against a real host. See Testing strategy.