القائمة
Switch model mid-session preserving thinking blocks. Used by /model command and by BRD §6.2 failover when the primary provider rate-limits or fails.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Six-phase ReAct loop (pre-check, thinking, self-critique, action, tool, post). Activates per-workflow via the thinking_level knob in config/workflows.yaml. Replaces standard ReAct for workflows where independent verification of the reasoning improves outcomes.
Mines completed sessions for repeating {tool sequence → outcome} tuples. Scores by frequency × success_rate × novelty. High-scoring tuples become "instincts" — candidate skill seeds — in instincts/pending/.
Walk-back algorithm for the Spec-Auditor subagent. Given a failure at phase N, identifies the earliest phase whose spec, if tightened, would have prevented the failure. Proposes a surgical amendment.
Sessions stored as trees (not lists). /fork creates a branch from any point; /tree navigates; /branch labels a path; /export produces HTML for review. All branches live in one session file under sessions/<project>/<session_id>.json.
Progressive context refinement for subagents. Don't dump everything; retrieve in passes. Start with a list of file paths and one-line abstracts; load full content only when the abstract proves insufficient.
Create a Business Requirements Document through Socratic five-dimension dialogue with the human. First step of the SDLC pipeline, before spec/design/build. Supports greenfield projects or single-feature additions.
| name | cross-provider-handoff |
| description | Switch model mid-session preserving thinking blocks. Used by /model command and by BRD §6.2 failover when the primary provider rate-limits or fails. |
| when_to_use | ["user invoked /model to switch providers","primary provider returned 429 / 503 / network error","workflow declared failover[] in config/workflows.yaml and the primary is unavailable"] |
| brd_ref | §6.2 |
Pi (@mariozechner/pi-ai) supports mid-session model switching across Anthropic, OpenAI, Google, Bedrock, Vertex, Azure, Ollama, vLLM, LM Studio. Thinking blocks are auto-converted between provider formats. Cost-per-token is tracked per provider against Pi's pricing table.
| From → To | Thinking block handling |
|---|---|
| Anthropic → OpenAI | thinking content blocks → reasoning content (where supported) or omitted |
| OpenAI → Anthropic | reasoning_content → thinking blocks |
| Anthropic → Google | thinking → comment headers in the next user message |
| Anthropic → Bedrock | passes through (same Claude model family) |
The conversion is lossy in one direction: thinking content tagged for redaction by Anthropic policy cannot be passed verbatim to a third-party provider. The handoff logic drops these blocks; the agent may briefly lose reasoning continuity, but the conversation continues.
compactor and Haiku is up. Don't fail over compaction; it's cheap to retry on the primary.primary → failover[0] fails, surface the error to the orchestrator rather than continuing to failover[1] in the same turn.Wraps earendil-works/pi/packages/pi-ai/src/providers/ (MIT). See BRD §6.2 and §10.4 §6.2.