Runtime Selection

Set host_python to the current host's explicit absolute interpreter path (not a PATH lookup). The resolver below is stdlib-only, and host_python runs it and nothing else. Resolve the configured interpreter before any other toolkit command:

"{host_python}" "{speaker_toolkit_root}/skills/shownotes-publisher/scripts/resolve-runtime.py"

Input: optional --database <path>, defaulting to the documented vault locator ~/.claude/rhetoric-knowledge-vault/tracking-database.json.

Output: one JSON object on stdout. On ok: true (exit 0), set python_path and vault_root from it (the root is the database locator's directory, never config.vault_root); every later Python toolkit command in this skill and its references runs on "{python_path}". Shell scripts keep the bash invocation their step shows.

On ok: false (exit 1) with reason_code: database_generation_conflict, another writer changed the database mid-read: run the resolver once more before anything else. On any other ok: false, stop publishing and invoke Skill(skill: "vault-ingress") with this repair-only handoff as context: runtime_repair: {reason_code: "{reason_code}", lane: "outline", database: "{database}"}, where database is the failure's absolute locator. That entry runs only the repair the code names, against that database alone, and returns without starting an ingest. When it returns, run the resolver once more with --database "{database}" and continue with Step 1 on ok: true. When the repair reports nothing it could fix, or the resolver refuses again, finish here and report the reason_code. Never fall back to whichever python3 is on PATH.

The checks the resolver composes, and its reason codes, are its own contract: see skills/shownotes-publisher/scripts/resolve-runtime.py (module docstring).

The same resolver also supports focused presentation tasks through presentation runtime context. Their optional read-only/core modes do not change this publisher's default: owner-current database and outline runtime before publishing.