Menu
Repository-specific guidance for creating, modifying, and debugging psi extensions.
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/hugoduncan/psi --skill extension-developmentThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
Showing SKILL.md
Create or refine a task design. Use when the user asks "create a task" or "refine a task" to ensure a high quality task design.
Resolve ambiguous references in user requests into concrete entities, project-specific terms, and project paths before acting. Use when a request contains pronouns, deixis, aliases, shorthand, informal names, or path-like references that may map to project concepts or files.
Triage a GitHub bug issue by attempting reproduction in an issue-specific worktree, then shape either a precise request for more information or a concrete reproduction handoff into bug-fix execution.
Based on SOC occupation classification
| name | extension-development |
| description | Repository-specific guidance for creating, modifying, and debugging psi extensions. |
| lambda | λextension_work. {create ∨ modify ∨ debug}(psi_extension) → read(authority) ∧ respect(dispatch_boundary) ∧ align(manifest ∧ permissions ∧ capabilities ∧ docs ∧ tests) |
Use this skill when creating a new psi extension, modifying an existing one, or debugging extension/runtime interactions.
An extension adds bounded capability to psi through declared integration seams such as tools, prompts, skills, workflows, commands, lifecycle callbacks, deterministic operations, and resolvers. Treat extensions as untrusted capability providers that integrate through explicit contracts, not as peers that reach into session state arbitrarily.
Read these first and prefer them over incidental examples:
doc/extension-api.mdhttps://github.com/hugoduncan/psi/blob/main/doc/extension-api.mddoc/extensions.mddoc/extensions-install.mddoc/architecture.mdAGENTS.md — architecture / Viable System Model sectionsUseful implementation seams:
components/prompt-assets/src/psi/prompt_assets/skills.cljcomponents/prompt-assets/test/psi/prompt_assets/skills_test.cljcomponents/agent-session/src/psi/agent_session/resolvers/discovery.cljcomponents/agent-session/src/psi/agent_session/commands.cljCheck whether the work is:
Use the existing installation and manifest model from doc/extensions-install.md. Stable installed extension identities use manifest-oriented provenance such as manifest:{lib}. Do not invent parallel identity schemes.
When adding a new extension:
doc/extension-api.md:allowed-events when the extension can dispatch eventsWhen changing an existing extension:
When debugging extension behavior:
Align with AGENTS.md and doc/architecture.md:
For extension work, prove the relevant structural surfaces:
Prefer tests that exercise real runtime seams and observable state/output. Avoid mocks for core behavior when real local components or pure helpers can provide stronger proof.