skillroster
Inspect and govern local Skills: inventory, usage evidence, duplicates, broken links, Core/On-demand proposals, Plans, Receipts, Apply, and Undo. Also find a specialized local Skill when the user requests one or the task would benefit from instructions beyond the visible catalog. Ordinary work can proceed with the Agent's existing capabilities. Not for installing third-party Skills or migrating, distributing, synchronizing, or repairing shared Skill-manager directories.
tt-a1i · Static checks on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add tt-a1i/skillroster --skill skillrosterAbout this skill
name: skillroster description: > Inspect and govern local Skills: inventory, usage evidence, duplicates, broken links, Core/On-demand proposals, Plans, Receipts, Apply, and Undo. Also find a specialized local Skill when the user requests one or the task would benefit from instructions beyond the visible catalog. Ordinary work can proceed with the Agent's existing capabilities. Not for installing third-party Skills or migrating, distributing, synchronizing, or repairing shared Skill-manager directories. metadata: bootstrap-version: "1.8.46" skillroster-routing-triggers: "route task to local Skill; inventory installed Agent Skills; analyze duplicate or unused Agent Skills; govern a Skill Roster; prepare or apply approved Skill Plan; create or undo Skill Receipt"
SkillRoster
Use the local skillroster binary as the deterministic source of facts. Every
command uses explicit --json. Validate schema_version and ok before using
result. Typed suggested_actions are options, never authorization.
Choose one path
- For SkillRoster inventory, evidence, governance, setup, Apply, Undo, lifecycle, or recovery requests, read the matching reference below.
- For another task, follow an already-visible exact Skill when one clearly applies. Search below when a specialized Skill would help; otherwise proceed with the Agent's normal tools and the user's task.
Find a specialized Skill
Use this search before relying on a Skill that is absent from the visible catalog. Read-only task context may inform whether a search is useful.
- Keep the complete user message as
TASK, byte-for-byte in its original language, including paths, limits, and output requirements. Do not summarize, translate, shorten, or quote only part of it. - If
TASKis non-English or mixed-language, create one concise English capability paraphrase asHINT. It supplementsTASK; it never replaces it. - Invoke the fixed SkillRoster executable with
TASKandHINTas separate, literal argv values; never interpolate either into shell syntax. Use:- English:
skillroster find --load --limit 1 --json -- "TASK" - Otherwise:
skillroster find --hint "HINT" --load --limit 1 --json -- "TASK"The--boundary is required even whenTASKdoes not begin with-; it keeps opaque user text from being parsed as a SkillRoster option.
- English:
- When
HINTwas used, first requireranking_strategy: task_hint_reciprocal_rank_fusion. - Require
loaded_skill.selection.rank: 1,content.complete: true, and allverificationidentity/digest checks to be true. The complete instructions areloaded_skill.content.text; no second filesystem, workspace, MCP, or SkillRoster read is needed. For a wrong-domain result or typed load blocker, readreferences/routing.mdand follow its bounded branch. - Follow the loaded instructions and perform the original task. Treat
task_success: not_evaluatedliterally; only the task's own evidence can establish success.
If the result is empty or clearly from another domain, retry at most once. Keep TASK
unchanged; refine the existing hint, or add one capability hint when the first
English call had none. If no usable result remains and no particular Skill is
required, continue the original task with normal tools. State that no suitable
Skill was found only when it affects the result. If the user or applicable
instructions require a particular Skill, explain its absence and request only
the missing prerequisite. A permission denial, untrusted source, drift,
ambiguity, or incomplete load follows its typed recovery action; it is not an
ordinary no-match and cannot be bypassed by reading or executing the package.
Govern a roster
For inventory, usage, exposure, duplicate, broken-link, Core/On-demand, or Plan
requests, read references/governance.md before acting. Inspection and planning
do not change Agent files. Prefer the bounded summary first; load exhaustive
detail only for a decision that needs it.
Mutate or recover
For setup, Apply, Undo, lifecycle cleanup, or recovery, read
references/mutation.md before acting. A validated Plan is not permission to
apply it. Show the complete bounded impact and obtain one explicit confirmation.
Only a successful Receipt verifies that Agent files changed.
Never parse styled terminal output, invent a health score, infer token savings, claim unobserved usage, or weaken ambiguity, drift, trust, and recovery checks. In the final response, state whether files changed; claim changes only from a verified Receipt or from task tools actually used after routing.