yuzu482/yhwh/pi-dispatch

Claude Code and agent skill by yuzu482/yhwh

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Governed local Pi MCP gateway for Tifereth's default lower-agent execution/probe layer; not Codex desktop task management.

Install

npx vskill@latest install yuzu482/yhwh/pi-dispatch

SKILL.md

Pi Dispatch

Use http://127.0.0.1:17331/mcp. Tifereth owns scope, routing, integration and acceptance. Use bounded Pi tasks; built-in subagents need explicit user request. No recursion.

Before dispatch

  • Check list_capabilities and read pi-routing for scope, acceptance and adaptive thinking. Listed tuples are allowlists, not live access; only an authorized probe_model tests access.
  • Workers use openai-codex / gpt-6-luna; Geburah reviewer uses claude-code-cli / claude-sonnet-5 (medium default; high/xhigh by review complexity), access:none and task.reviewPacket. The host may explicitly select the separate anthropic API route. No automatic route substitution; exact role bindings are in pi-routing.
  • Grant least privilege (none, read, verified sandboxed workspace-write) and allowed absolute cwd; writes are proposals. Pi roots only; credentials runtime-side, never prompts/tasks/packages. See pi-routing for conditional gateway mechanics.

Submit and wait

Use AGENTS.md tiers; task-tiers for boundaries. Send tier/five semantic flags; gateway counts, rejects unclassified writes. T0 may dispatch synchronously; otherwise submit asynchronously with stable request/parent IDs, exact route/cwd/access/profile and role/objective/acceptance. No raw prompts, commands, environment or tool lists; schemas: pi-contracts.

Only effective T2 writes require a task-tiers receipt; T0/T1 dispatch and host verification do not. Memory/graph tools retain their receipts. Never poll repeatedly; wait for event, deadline or user request under pi-results. Reconcile uncertain writes using the same ID; stop on idempotency_key_reused or in_doubt.

Accept and conditional mechanics

Accept only terminal completed with outcome.ok=true, correct route, valid format/role, full evidence and cleanup. See pi-results for retrieval and validation.

T2 uses v2 links; never forge upstream or disguise dependencies. Upgraded known-scope: primary plan → approved pre-review → Chesed → bound record_host_verification → post-review. T0/T1/T2 use real host outputs/exit codes; Netzach needs unavailable host execution or a concrete independent check. T1 non-diff ≤8 KiB / total ≤24 KiB; split by commit under pi-review; approval/conditional stops. Elastic quota/activation: pi-review. Follow actual admitted stages before upgrade; details: task-tiers, pi-contracts, pi-review.

Choose the lowest justified whole-goal tier; explain escalation with TIER_REASON=. Authorized host-diagnosed repairs need no renewed permission and never reset reviews. Validate complete material/bytes before launch; no truncation/path-only review. Corrections need current artifact evidence; label reused coverage.

LSP uses direct lsp_request, not model delegation; inspect status/backend/raw result. Degraded is reduced evidence; unavailable/failed fail acceptance; see pi-lsp.

Optional transports require operator config; digest changes fail closed. No auto-switch, lowered thinking or credential fallback; authorized probes only. Full credential rules: pi-auth.

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
52
Repository forks
2
7-day trend score
25

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/yuzu482/yhwh/pi-dispatch/badge)](https://verified-skill.com/skills/yuzu482/yhwh/pi-dispatch)

Something wrong with this skill or its page? Tell us.