feat: add Pi and OMP trial stacks

[kentwelcome]

Summary

• Add first-class Pi and OMP trial stacks while keeping Claude Code and Codex as the plugin hosts.
• Normalize Pi and OMP JSON events into the existing canonical trace format and pin decision extraction to the selected stack and model.
• Document host-specific live behavior, including the honest no-dispatch Codex fallback, and add deterministic regression coverage.

Closes DRC-4282

Test plan

• docker run --rm -v "$PWD:/mnt" -w /mnt mvdan/shfmt:v3.14.0 -d -i 2 -ci .
• uvx ruff@0.16.5 format --check --diff .
• bash tests/hooks-test.sh
• python3 plugin/skills/behavior-diff/scripts/decisions.py --check
• bash tests/live-report-contract.sh
• bash tests/release-workflow-test.sh
• Independent read-only review: APPROVE
• Paid live Pi and OMP smoke runs were not run. They require explicit cost approval, and the installed Pi currently cannot start under Node 20.20.1.

[Copilot]

🟢 Approval recommended

The Pi/OMP support is implemented end-to-end (runner, trial adapter, decision extractor, docs) with deterministic contract coverage and no issues found in the changed logic.

Pull request overview

This PR extends Behavior Diff to support Pi (pi) and OMP (omp) as first-class trial stacks (while keeping Claude Code and Codex as the only plugin hosts). It adds stack-specific headless CLI execution + normalization into the existing canonical trace.jsonl format, pins decision extraction to the selected stack/model, and updates docs and deterministic contract tests to lock the new behavior.

Changes:

• Add pi/omp support to the headless runner (behavior-diff.sh) and per-trial adapter (run-trial.sh), including strict --model requirements for Pi/OMP.
• Extend decision extraction (decisions.py) with Pi/OMP extractor runners and self-check coverage for pinned extractors and model propagation.
• Update live-skill host dispatch docs, public README wording, plugin manifest descriptions, and deterministic contract tests to cover the new stacks.

File summaries

File	Description
tests/live-report-contract.sh	Adds contract assertions for ownership wording, Pi/OMP stack/model mentions, Codex no-dispatch behavior, and README/manifest wording.
tests/hooks-test.sh	Adds deterministic shell coverage for new stacks, including normalization expectations and model/CLI validation paths.
README.md	Clarifies plugin hosts vs trial stacks and adds a support matrix covering Claude/Codex/Pi/OMP surfaces.
plugin/skills/behavior-diff/SKILL.md	Documents ownership split (skill judgment vs script mechanics) and adds explicit Pi/OMP stack + exact-model guidance.
plugin/skills/behavior-diff/scripts/run-trial.sh	Implements Pi/OMP CLI branches and shared JSONL normalizer into canonical trace.jsonl.
plugin/skills/behavior-diff/scripts/decisions.py	Adds Pi/OMP extractors, enforces model requirement for pinned Pi/OMP, and expands self-check coverage.
plugin/skills/behavior-diff/scripts/behavior-diff.sh	Accepts pi/omp, enforces explicit model for them, improves unsupported-stack messaging, and defaults extractor stack/model for Pi/OMP runs.
plugin/skills/behavior-diff-live/SKILL.md	Updates live dispatch instructions for Claude Code, OMP (task batch), Pi (no dispatch → headless), and Codex no-dispatch fallback.
plugin/.codex-plugin/plugin.json	Updates plugin description to list trial stacks (claude, codex, pi, omp).
plugin/.claude-plugin/plugin.json	Updates plugin description to list trial stacks (claude, codex, pi, omp).
plans/2026-09-03-drc-4282-pi-omp-trial-stacks.md	Adds a detailed implementation plan and acceptance criteria for DRC-4282.
bin/behavior-diff	Updates CLI usage comment to include `pi

Review details

• Files reviewed: 11/12 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.