pi-playbooklisted

# pi Playbook `pi` is a coding agent CLI with a flag surface that maps unusually cleanly onto Skill Heaven's mechanisms — it can evict and readmit skills natively, so postures compose without any config-directory tricks. Verified against **pi 0.83.0**. Re-check `pi --version` before trusting any dose measurement; a flag's behaviour has changed across versions before (see §5). > **Rule 0 — run pi in a herdr pane, never through your Bash tool.** The operator must be able to > see which model you invoked. A probe the operator could not see is not evidence. Use > `pane send-text` with a trailing newline **inside** the quotes — **not** `pane run`, which > splits argv on whitespace and destroys quoted prompts: > > ```bash > herdr pane send-text "$PROBE_PANE" 'pi --model openai-codex/gpt-5.6-luna:low -p --no-session "probe text" > ' > herdr pane read "$PROBE_PANE" > ``` > **Building the door?** Read `packages/core/skills/harness-door-pattern/SKILL.md` first — it > carries the suppression-mechanism taxonomy, the probe methodology, and the traps. This playbook > is the harness-specific detail underneath it. --- ## 1. Model selection ```bash pi --list-models # all models pi --list-models luna # fuzzy search ``` Models are addressed as `provider/id`, with an optional `:thinking` suffix: ```bash pi --model openai-codex/gpt-5.6-luna:low "..." pi --model openai-codex/gpt-5.6-luna:max "..." ``` Session guinea pig: **`openai-codex/gpt-5.6-luna:low`**