SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsinstall-hve-core-rpiv

install-hve-core-rpiv

Agents & models

View repository

Updated 10-16 06:19

About this skill

install-hve-core-rpiv

You install a local, skill-compatible projection of the HVE Core RPI/RPIV workflow.

HVE Core is authoritative. This skill does not fork RPIV and does not maintain generated task-* skills as catalog source. Each run acquires the current HVE Core default branch, reads the RPIV prompt/agent assets, generates local SKILL.md wrappers, records the resolved source commit, and reports how to update or remove the projection.

Hard rules

  1. Ask for consent before writing files.
  2. Default to a project-local destination: .agents/skills.
  3. Do not install globally unless the user explicitly asks for a user/global destination.
  4. Prefer GitHub CLI (gh) for acquiring HVE Core when available; with gh, fetch the required files directly instead of cloning the repo. Fall back to plain git clone only when gh is unavailable or declined.
  5. Never modify the HVE Core source repository.
  6. Do not commit generated task-* skills into this tools catalog as maintained source.
  7. Do not claim the generated skills are exact HVE Core agents or prompts. They are skill-shaped projections.
  8. Do not create partial installs: if source acquisition or validation fails, leave the destination unchanged.

Generated skills

Generate exactly these v1 skills:

  • task-research
  • task-plan
  • task-implement
  • task-review

Do not generate an rpi orchestrator skill in v1.

Each generated skill must disclose:

  • It was generated by install-hve-core-rpiv.
  • HVE Core remains authoritative.
  • The source repository, source ref, and resolved source commit.
  • The upstream prompt and agent files used.
  • Rerun install-hve-core-rpiv to update instead of hand-editing generated content.
  • Skill runtimes may not preserve HVE Core agent picker, subagent, handoff, model-selection, or prompt-dispatch semantics.

References

Read these before writing:

  • references/source-acquisition.md
  • references/generated-skill-contract.md
  • references/manifest-and-drift.md

Install flow

  1. Explain the intent:
    • HVE Core already has RPIV as agents/prompts/instructions.
    • This skill projects those assets into local supported skills for environments that scan SKILL.md folders.
    • Generated skills are derived artifacts, not a fork.
  2. Ask for consent to inspect the environment and write local files.
  3. Ask for destination, defaulting to .agents/skills.
    • Offer .pi/skills only when the user is specifically testing Pi.
    • Allow a custom path.
    • Warn before any user/global destination.
  4. Acquire the required HVE Core RPIV files using references/source-acquisition.md.
  5. Validate required upstream RPIV source files exist.
  6. Resolve and display the HVE Core source ref and commit.
  7. Read any existing manifest from the destination.
  8. Check for unmanaged collisions and managed-file drift before writes.
  9. Render generated skills to a temporary staging directory.
  10. Validate staged output.
  11. Atomically copy staged managed files into the destination.
  12. Write or update the manifest.
  13. Clean up temporary file caches, fallback clones, and staging directories or report the exact cleanup path if cleanup cannot complete.
  14. Print a concise final report.

Destination safety

Before writing:

  • Normalize the destination path.
  • Reject path traversal or writes outside the selected destination root.
  • If the destination is inside a git worktree and already contains uncommitted generated files, stop and ask before changing them.
  • If an existing target skill is not listed in the manifest as managed by this installer, refuse to overwrite unless the user explicitly confirms.
  • If a managed file hash differs from the manifest, explain the drift and ask before overwriting.

Validation checklist

After staging and before reporting success, verify:

  • Destination root exists or can be created.
  • Four generated skill folders exist.
  • Each generated skill contains SKILL.md.
  • Each SKILL.md has name: and description: frontmatter matching the generated skill.
  • No agent-only keys leaked into generated frontmatter: agent, agents, handoffs, disable-model-invocation.
  • Each generated skill names its source prompt, source agent, source ref, and source commit.
  • The manifest exists and includes source_repo, source_ref, source_commit, destination, managed_skills, and managed_files.
  • Manifest hashes match staged files before install and installed files after install.
  • Rerunning at the same commit can report already up to date or revalidate without churn.
  • Network/auth failures from gh or git leave the destination unchanged.
  • Temporary file caches, fallback clones, and staging directories are removed or explicitly reported for cleanup.

Final report

On success, report:

  • Destination path.
  • HVE Core source ref and commit.
  • Generated skill paths.
  • How to invoke or discover the skills in the user's client.
  • How to rerun the installer to update from HVE Core.
  • How to roll back or uninstall using the manifest.
  • A clear ready message telling the user installation is complete and they should restart or reload their coding agent before invoking the generated task skills.

Use this completion wording, adjusted only for exact client names and paths:

RPIV local skill installation is complete. Restart or reload your coding agent so it discovers the new skills, then invoke the generated RPIV task skills.

On failure, report: