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, not mutually exclusive task owners. Modularity encapsulates reusable logic and authority routing; composability exposes ports through which selected capabilities consume the same task data and contribute decisions or evidence.
Each row describes ports, not an exclusive category.
| Capability | Consumes | Provides | Composition seam | Boundary |
|---|---|---|---|---|
bootstrap-project-context |
Repository state and persistence pressure | Context ownership migration | Composes with the affected concern and prose capabilities | Does not own ordinary implementation |
clarifying-contracts |
An unresolved user-owned semantic choice | An approved contract | Adds authority resolution to any mixed task | Does not resolve discoverable facts |
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 without changing its semantic owner | Does not activate for latency alone or prescribe agent micro-orchestration |
software-engineering |
Software contract and implementation evidence | Behavior, topology, testing, and delivery decisions | Composes with language and prose capabilities | Does not own host-language syntax |
julia-development |
Julia project, version, source, and renderer evidence | Julia-specific decisions and host conventions | Composes with software, structure, and Markdown capabilities | Does not own language-independent software topology |
structure-documentation |
Established durable meaning and the reader's task | A language-independent reading path | Composes with artifact-language and host capabilities | Does not decide meaning or syntax |
markdown-authoring |
A Markdown artifact and renderer facts | Markdown syntax and generic or GitHub rendering realization | Composes with structure and host extensions | Does not own prose semantics or host-specific extensions |
obsidian-cli |
Obsidian index or runtime state | Runtime-aware Obsidian operations | Composes with Markdown and structure only when content changes | Does not own ordinary filesystem editing |
Validate every published skill against the Agent Skills specification:
python3 -m unittest discover -s tests
gh skill publish --dry-runclarifying-contracts adapts Superpowers' brainstorming workflow and Matt Pocock's grilling1. obsidian-cli adapts and narrows Kepano's upstream Obsidian CLI skill.
- The Agentic Tend principles own the motivation and pressure for 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 development capability model illustrates composition without prescribing a fixed workflow.
- The organization roadmap tracks evidence and extension work spanning multiple tooling layers.
The repository is available under the MIT License. Each independently installable skill bundles its complete terms and any required upstream notices.