Skip to content

Collection Review

github-actions[bot] edited this page Sep 30, 2026 · 1 revision

Meta-skill collection review

Reviewed on 2026-09-19 against the 23 canonical package entrypoints, their responsibility boundaries, handoffs, selected implementation/provenance resources, and rendered sibling icons. This is a semantic collection review; it is not a model benchmark or a claim that every possible domain skill belongs here.

The repository owns only procedures directly about skills. Tool, framework, infrastructure and general execution packages belong in a separately authorized collection. Global package work is outside this repository completion. A package name alone does not prove this boundary: future additions need the same trigger, output and consumer review.

Package Decision and boundary rationale Icon decision
skill-authoring Retain the package-producing coordinator. Discovery, research, synthesis, design and evaluation each return a distinct input; the coordinator owns their handoffs and final package. Remove the arbitrary line rejection so complete ordinary examples remain together. Retain pencil: authors the candidate.
skills-discovery Retain qualified external-package discovery. It inspects the full relevant source corpus and rights; domain truth remains research's responsibility. Retain search lens.
skill-domain-research Retain the separate authoritative subject dossier. It reconciles process-owner practice, package claims and current external facts before design. Retain evidence matrix and lens, distinct from package discovery.
skills-synthesis Retain a contribution plan for at least two distinct useful packages. One source goes directly to design; mirrors never inflate the count. No merge is justified merely by shared tooling. Retain funnel: selects useful contributions.
skill-design Retain one responsibility/interface decision before implementation. Its preliminary scoping can precede research; the final design requires the reconciled evidence. Retain lightbulb.
skill-evaluator Retain candidate/baseline evaluation against frozen cases. It judges evidence and returns a verdict; it neither writes fixes nor decides publication. Retain checklist with result mark.
skill-evidence-collection Retain a bounded packet of observations for a later evaluator or lifecycle decision. Gathering evidence must not imply a passing verdict. Replace the checklist/checkmark with stacked documents; the former overlapped evaluation. Preserve source license and a rendered PNG.
skill-evolution Retain one evidence-supported package update and change record. It decides a supported revision; iterative measured search belongs to optimization. Retain update arrow and clock.
skill-optimization Retain measured candidate improvement with separate tuning, selection and final acceptance. No context reduction is accepted at the cost of required coverage. Retain rising trend.
skill-naming Retain a naming decision without filesystem edits. Cardinality and collision evidence can be requested separately from a full design or migration. Retain label.
skill-icon-design Retain one responsibility-specific asset. It consumes the package boundary and sibling visual set without redesigning the package's procedure. Retain palette.
skill-installation Retain an authorized consumer installation and receipt. Source identity, collision handling and rollback distinguish it from migration and publication. Setup remains explicitly unexecuted. Retain download arrow.
skill-migration Retain a package move with source/target/consumer transition evidence. A move does not authorize a new behavior or remove the source implicitly. Retain opposing arrows.
skill-publication Retain authorized channel publication and read-back receipt. Preparation and local checks cannot satisfy or authorize that external outcome. Retain upload arrow.
skill-lifecycle-review Retain a maturity/transition record from current evidence. Lifecycle judgments remain outside the derived catalog and do not modify packages. Retain review annotation.
skill-routing Retain a task-specific selection, ordered route, ambiguity or none. It reads a shortlist's actual entrypoints and setup readiness without executing setup. Retain branching route.
skill-security-review Retain one package's security/disclosure decision. It examines instructions, code and provenance; a general quality verdict belongs to evaluation. Retain shield.
skills-audit Retain collection-wide findings with frozen inventory and policy. The audit detects defects; it does not implement a refactoring plan. Retain shield and lens.
skills-catalog Retain derived inventory and optional cross-source/history projections. Catalogs remain metadata, not lifecycle approval, task tracking or automatic routing. Retain list.
skills-refactoring Retain a bounded collection reorganization plan. Every revision reassesses scope, names, overlaps, consumers and retain/merge/split/compact alternatives. Retain hierarchy.
skills-host-compatibility Retain the structural report for declared local aliases. It cannot infer native host discovery or install a host configuration. Retain converging paths.
skills-snapshot Retain content-addressed capture, verification and restore as one recovery lifecycle. Deduplication and retained historical manifests support rollback; migration applies the actual move separately. Include the bundled tests in the central check. Retain archive box.
skills-maintenance-scheduling Retain a bounded skill-maintenance cadence proposal, with a real executor and explicit authority. Generic scheduling and unattended mutation remain outside its responsibility. Retain calendar.

No package rename, merge or split is supported by the current triggers and outputs. Shared vocabulary is deliberate affinity, and the retained handoffs avoid duplicate implementation ownership. General domain examples in meta-skill instructions illustrate acceptance criteria; they are not embedded domain packages. Useful resources stay within their owning packages and remain available when that topic is selected.

All icons have distinct byte identities. The visual review compares metaphor, silhouette and small-size legibility as well as palette; color alone is not the distinction. Structural checks validate SVG safety, raster format and host asset references. These checks do not imply an accessibility certification.

Revisit this decision map when a package's primary output, consumer, authority or independently testable responsibility changes. Catalog synchronization checks the current inventory; this dated decision record does not replace it.

Clone this wiki locally