Skills

This repository owns Agentic Tend's reusable skills layer, dogfooded in Codex before publication.

Each top-level skill directory is independently discoverable through its SKILL.md entrypoint. The collection follows the open Agent Skills format: compatible agents discover concise metadata first and load full instructions and bundled resources only when task traits match. Agentic Tend's capability model owns the primitive ontology, its context ownership model owns placement and activation boundaries, and this README instantiates the current capability graph. Semantic ownership, activation, and mechanical enforcement remain distinct.

Skills are composable capabilities dispatched from task traits. Modularity encapsulates reusable logic and authority routing; composability exposes ports through which selected capabilities consume the same task data and contribute decisions or evidence.

Capability topology

Each row describes ports and the adjacent owner or route that completes the composition.

Capability	Consumes	Provides	Composition seam	Adjacent owner or route
bootstrap-project-context	Repository state and persistence pressure	Context ownership migration	Composes with the affected concern and prose capabilities	Ordinary implementation -> the affected concern
clarifying-contracts	An unresolved user-owned semantic choice	An established contract	Adds authority resolution to any mixed task	Discoverable facts -> inspection or evidence; realization -> the affected concern
multi-agent-evidence	Task data with independently formable, held-out, or falsifiable evidence	A human-visible plan, dynamic evidence ports, recursive feedback, and provenance-aware delivery	Places selected capability work in separate contexts while preserving its semantic owner	Capability meaning -> selected capability; action authority -> task data
scientific-literature-evidence	A literature question requiring a source corpus, attribution, version comparison, conflict resolution, or bounded absence search	Source roles, claim-level evidence, synthesis, version state, and search boundaries	Supplies literature evidence to scientific claim audit and durable research prose	Manuscript-claim validity -> scientific claim audit; bibliography realization -> artifact tooling
scientific-claim-audit	A specific scientific claim or objection and its controlling evidence	Claim-level support status, scope, findings, and unresolved gaps	Consumes formal, computational, or literature evidence and supplies scientific meaning to prose or delivery capabilities	Final prose -> structure documentation; artifact syntax -> artifact tooling
referee-response	An exact scientific referee comment, current manuscript state, and established evidence	A scoped response and manuscript-revision map	Supplies established meaning to prose organization and repository-specified artifact workflows	Contested validity -> scientific claim audit; manuscript or submission action -> its authorized workflow
software-engineering	Software contract and implementation evidence	Behavior, topology, testing, and delivery decisions	Composes with language and prose capabilities	Host syntax -> language capability; prose expression -> structure documentation
julia-development	Julia project, version, source, and renderer evidence	Julia-specific decisions and host conventions	Composes with software, structure, and Markdown capabilities	Language-independent topology -> software engineering; prose -> structure documentation
structure-documentation	Established durable meaning and the reader's task	A language-independent reading path	Composes with artifact-language and host capabilities	Meaning -> affected concern; syntax -> artifact or host capability
markdown-authoring	A Markdown artifact and renderer facts	Markdown syntax and generic or GitHub rendering realization	Composes with structure and host extensions	Prose meaning -> structure documentation; host extensions -> host capability
obsidian-cli	Obsidian index or runtime state	Runtime-aware Obsidian operations	Composes with Markdown and structure when content changes	Ordinary files -> filesystem or Markdown tooling; prose -> structure documentation

Validate

Validate every published skill against the Agent Skills specification:

python3 -m unittest discover -s tests
gh skill publish --dry-run

Provenance

clarifying-contracts adapts Superpowers' brainstorming workflow and Matt Pocock's grilling¹. obsidian-cli adapts and narrows Kepano's upstream Obsidian CLI skill.

The scientific capabilities adapt distinct parts of TeXRA's MIT-licensed workflows:

• scientific-claim-audit retains validity discipline from manuscript-review and inline-paper-critic.
• scientific-literature-evidence refactors literature-search around source roles, versions, and bounded scientific synthesis.
• referee-response adapts response-delivery constraints from manuscript-review and writing-commenter, while leaving prose organization to structure-documentation and artifact realization to repository instructions and current tools.

The physics-journal specialization of referee-response follows current APS and IOP author guidance; general online writing skills remain design references rather than procedural authorities.

See also

• The Agentic Tend principles own pressure-driven behavior and its specialization to persistent structure.
• The presentation model owns progressive disclosure for human review.
• The multi-agent evidence model owns the effective theory for observable agent ports, feedback topology, and evidence separation.
• The capability-composition examples illustrate composition without prescribing a fixed workflow.
• The organization roadmap tracks evidence and extension work spanning multiple tooling layers.

License

The repository is available under the MIT License. Each independently installable skill bundles its complete terms and any required upstream notices.

Footnotes

1. Matt Pocock, grilling: one-question-at-a-time clarification of plans and decisions. ↩