pi-cli-runtime
agents365-ai/365-skills
Internal helper for invoking pi-companion runtime from Claude Code subagents.
What is pi-cli-runtime?
This is an internal skill for the pi:pi-companion-forwarder subagent only. It provides a standardized contract for calling the Pi runtime via a single task invocation, with rules for argument handling, model selection, effort levels, and resume behavior.
- Invokes pi-companion runtime via node script with task command
- Handles routing flags (--background, --wait, --resume, --fresh, --model, --effort)
- Strips execution-control flags before forwarding to Pi
- Supports resume-last for continuing previous runs
- Passes through model and effort parameters to Pi
- Returns Pi stdout unchanged without post-processing
How to install pi-cli-runtime
npx skills add https://github.com/agents365-ai/365-skills --skill pi-cli-runtime- Must be used only inside pi:pi-companion-forwarder subagent
- Requires CLAUDE_PLUGIN_ROOT environment variable set
- Requires pi-companion.mjs script at ${CLAUDE_PLUGIN_ROOT}/scripts/
How to use pi-cli-runtime
- 1.Call node script with task command and raw arguments
- 2.Strip routing flags (--background, --wait, --resume, --fresh) before task text
- 3.Pass --model through unchanged if user specified it
- 4.Pass --effort through unchanged if user specified it
- 5.Add --write by default unless user requests read-only behavior
- 6.Use --resume-last for resume requests, fresh task for --fresh requests
- 7.Return stdout exactly as-is without inspection or modification
Use cases
- Forwarding user rescue requests to Pi runtime from Claude Code
- Resuming interrupted Pi tasks with --resume-last
- Selecting specific models or effort levels for Pi execution
- Performing write-capable fixes with default --write flag
- Routing read-only diagnosis or research requests to Pi
- Claude Code subagent developers
- Internal pi:pi-companion-forwarder implementations
- Teams building rescue or code-fix workflows
pi-cli-runtime FAQ
Only inside the pi:pi-companion-forwarder subagent. It is not user-invocable and should not be called from other contexts.
No. The forwarder must invoke task once and return stdout unchanged. Do not inspect the repo, solve tasks yourself, or add independent analysis.
Strip both from the task text. Use --resume-last for resume requests and a fresh task invocation for --fresh requests.
No. Leave both unset unless the user explicitly requests them. Only add --model or --effort when the user asks for a specific value.
Return nothing. Do not attempt recovery, inspection, or follow-up work.
Full instructions (SKILL.md)
Source of truth, from agents365-ai/365-skills.
name: pi-cli-runtime description: Internal helper contract for calling the pi-companion runtime from Claude Code user-invocable: false
Pi Runtime
Use this skill only inside the pi:pi-companion-forwarder subagent.
Primary helper:
node "${CLAUDE_PLUGIN_ROOT}/scripts/pi-companion.mjs" task "<raw arguments>"
Execution rules:
- The rescue subagent is a forwarder, not an orchestrator. Its only job is to invoke
taskonce and return that stdout unchanged. - Prefer the helper over hand-rolled
git, direct Pi CLI strings, or any other Bash activity. - Do not call
setup,review,adversarial-review,status,result, orcancelfrompi:pi-companion-forwarder. - Use
taskfor every rescue request, including diagnosis, planning, research, and explicit fix requests. - You may use the
pi-promptingskill to rewrite the user's request into a tighter Pi prompt before the singletaskcall. - That prompt drafting is the only Claude-side work allowed. Do not inspect the repo, solve the task yourself, or add independent analysis outside the forwarded prompt text.
- Leave
--effortunset unless the user explicitly requests a specific effort. - Leave model unset by default. Add
--modelonly when the user explicitly asks for one (e.g.deepseek-v4-pro,deepseek-v4-flash). - Default to a write-capable Pi run by adding
--writeunless the user explicitly asks for read-only behavior or only wants review, diagnosis, or research without edits.
Command selection:
- Use exactly one
taskinvocation per rescue handoff. - If the forwarded request includes
--backgroundor--wait, treat that as Claude-side execution control only. Strip it before callingtask, and do not treat it as part of the natural-language task text. - If the forwarded request includes
--model, pass the model id through totaskunchanged. - If the forwarded request includes
--effort, pass it through totask. - If the forwarded request includes
--resume, strip that token from the task text and add--resume-last. - If the forwarded request includes
--fresh, strip that token from the task text and do not add--resume-last. --resume: always usetask --resume-last, even if the request text is ambiguous.--fresh: always use a freshtaskrun, even if the request sounds like a follow-up.--effort: accepted values areoff,minimal,low,medium,high,xhigh,max. The aliasnonemaps tooff.task --resume-last: internal helper for "keep going", "resume", "apply the top fix", or "dig deeper" after a previous rescue run.
Safety rules:
- Default to write-capable Pi work in
pi:pi-companion-forwarderunless the user explicitly asks for read-only behavior. - Preserve the user's task text as-is apart from stripping routing flags.
- Do not inspect the repository, read files, grep, monitor progress, poll status, fetch results, cancel jobs, summarize output, or do any follow-up work of your own.
- Return the stdout of the
taskcommand exactly as-is. - If the Bash call fails or Pi cannot be invoked, return nothing.
Related skills
More from agents365-ai/365-skills and the wider catalog.

pi-prompting
Guidance for composing reliable prompts to Pi (DeepSeek by default) for coding, review, diagnosis, and research.

pi-result-handling
Internal guidance for presenting Pi helper output back to the user

plantuml-skill
Generate UML diagrams, flowcharts, and architecture charts as PNG/SVG via Kroki API—no local install required.

semanticscholar-skill
Search academic papers, citations, and authors via Semantic Scholar API with structured workflows.

target-prioritization
Prioritize drug targets from ranked gene lists using multi-source evidence (UniProt, OpenTargets, PubMed, HPA, ChEMBL) and composite scoring.

tldraw-skill
Generate hand-drawn whiteboard-style diagrams as .tldr JSON, export to PNG/SVG with tldraw-cli.