Presentation Runtime and Task Context

Use this shared bootstrap for authoring, building, publishing coordination, and CFP work. Select the action before opening vault documents. A forwarded task uses its destination skill's bootstrap; forwarding alone needs no vault load.

Resolve the runtime

Set host_python to the host's explicit absolute Python interpreter. The resolver is stdlib-only; it is the only toolkit command run under the host interpreter.

"{host_python}" "{speaker_toolkit_root}/skills/shownotes-publisher/scripts/resolve-runtime.py" --read-only --lane outline

Use --lane core for CFP/catalog work and standalone deck export. Omit --read-only immediately before publishing/tracking side effects: that requires the owner-current database. Local authoring, deck builds/edits/exports and resource extraction use the read-only compatibility mode; it grants no tracking-write permission. A catalog Markdown edit does not mutate the tracking database.

Input: optional --database <path>; otherwise the documented vault locator ~/.claude/rhetoric-knowledge-vault/tracking-database.json. On exit 0, take python_path, vault_root, database, database_sha256, database_state, and lane from the emitted JSON. Every later toolkit Python command uses that configured interpreter. Do not derive the root from stale config.vault_root, the working directory, or a PATH interpreter.

The resolver composes strict owner reads under the host and configured interpreters and requires matching bytes/generation. Its module docstring owns the status, runtime-probe and failure contracts. It never migrates the database.

On database_generation_conflict or database_mismatch, repeat discovery once. For other failures, invoke Skill(skill: "vault-ingress") with repair-only runtime_repair: {reason_code, lane, database} from the refusal, using the selected lane. Read the repair contract. Re-run against the same database after repair. A second refusal ends the task with the actionable diagnostic; never start an ingest or choose another Python.

Load only the selected context

All paths below are under the resolved vault root except talk-owned files. A missing required input blocks that action; an unrelated missing file does not. Project the profile before loading it into agent context:

"{python_path}" "{speaker_toolkit_root}/skills/presentation-creator/scripts/read-task-profile.py" "{vault_root}/speaker-profile.json" --task "{profile_task}"

Choose profile_task from the selected action below. Exit 0 emits {task, profile_available, profile} with selected configuration only. The script's TASK_FIELDS owns the field selection. It neither validates the owner profile nor authorizes historical claims; it never emits the history/badge branches. Owner validators and guardrail/history scripts still receive the original profile path. Read only the returned projection as context unless a later phase explicitly requires an authorized history source. Add --allow-missing for summary-only CFP/authoring or an export whose format is already supplied; a present invalid profile always refuses. For catalog-only reads, no profile projection is needed.

Use tasks authoring, cfp, catalog, build-pptx, build-presenterm, edit, export, resources, qr, delivery, or publishing as appropriate. During architecture's visual decisions, also select build-pptx for its visual/template configuration. Do not project publishing data for an isolated CFP or export.

For a first catalog, an absent sessions-catalog.md means no saved entries yet. Create it only when an approved catalog update is requested. Authoring/CFP work can use summary-only mode when a speaker profile has not been generated; do not invent structured history or configuration. Ask for missing task-specific choices. The prose summary is voice/nuance; the profile is structured configuration. Before making historical pattern claims, read pattern-history-authorization.md and run its status command. CFPs, exports and QR codes do not need that lane. Compare profile generated_date with summary Last updated only when both are loaded; warn on a stale profile without disabling independent configuration.

Read an outline as a non-owner

"{python_path}" "{speaker_toolkit_root}/skills/presentation-creator/scripts/outline_schema.py" --emit-json path/to/outline.yaml

Exit 0 emits the validated current outline JSON. Non-zero refuses without editing or migrating it. Invoke Skill(skill: "presentation-creator") with the exact outline path and diagnostic, then retry the original action after repair. Do not re-parse YAML to bypass a refusal.

Persist through the owner

The owner compatibility contract defines readable and writable generations.

Read the owner mutation contract before tracking writes. Re-resolve without --read-only before paired effects, re-read their current inputs, dry-run the typed mutation, inspect its changes, apply against the reported SHA, and strictly re-read. A successful read-only bootstrap or a previous run's hash is not write authority.