Manage Personas Skill

Persona resolution for the persona / ref / profile identity model. The resolve verb computes the transitive closure of a persona's composition DAG and emits one flat, deduped skills[] that a dispatch site passes as the execution-context's explicit skills[]. This is deterministic resolution — analogous to architecture resolve and manage-config resolve-recipe — not new runtime authority and not nested skill loading.

Enforcement

Execution mode: Single-verb script-executor — call resolve and route on the returned TOON status. Do not improvise additional verbs or arguments.

Prohibited actions:

• Do not hardcode a persona↔profile table — the persona's profiles: frontmatter is the binding source of truth.
• Do not hardcode a persona↔composition table — the persona's composes: frontmatter is the binding source of truth.
• Do not load composed personas via nested skill loading — composition is flattened by this resolver and carried in the explicit skills[].

Constraints:

• The composition graph MUST be a DAG; the resolver detects and rejects cycles (status: error, error: composition_cycle).
• The base persona-plan-marshall-agent is always included, unconditionally — it is never read from composes:.

What resolve does

Given --persona-key {bundle:persona} and optional --domains a,b,c, the resolver reads the persona's SKILL.md frontmatter and produces a flat, deduped skills[] by unioning, in deterministic order:

1. Base — always plan-marshall:persona-plan-marshall-agent (unconditional; same guarantee as the current foundational base load).
2. Direct composition — every bundle:skill notation in the persona's composes: frontmatter list (ref-* concerns and, for meta personas, other persona-* skills).
3. Recursive composition — for each composed persona-*, the transitive closure of its composes: and profiles: resolution (DAG walk; cycles rejected).
4. Profile × domain — for each profile in the persona's profiles: frontmatter list, the profile × {domains} domain skills resolved via the Extension API (manage-config resolve-domain-skills --domain {domain} --profile {profile}), for every domain in --domains. When --domains is omitted, no profile×domain skills are merged (the resolver emits base + composition only).

The persona's frontmatter (profiles: + composes:) is the sole binding source of truth — there is no hardcoded table anywhere in the resolver.

Output

status: success
persona_key: plan-marshall:persona-implementer
skills[N]:
  - plan-marshall:persona-plan-marshall-agent
  - plan-marshall:ref-code-quality
  - ...

On error: status: error with an error discriminator (persona_not_found, not_a_persona, composition_cycle, composed_persona_not_found).

Canonical invocations

The canonical argparse surface for manage_personas.py. The plugin-doctor analyzer (_analyze_manage_invocation.py) reads this section as source-of-truth for the manage-invocation-invalid and missing-canonical-block rules. Consuming docs xref this section by name instead of restating the command inline. See `pm-plugin-development:plugin-script-architecture` cross-skill-integration.md § "Script invocation in documentation".

manage-personas — resolve

python3 .plan/execute-script.py plan-marshall:manage-personas:manage_personas resolve \
  --persona-key PERSONA_KEY [--domains DOMAINS]

--persona-key is the bundle:skill notation of the persona to resolve (e.g. plan-marshall:persona-implementer). --domains is an optional comma-separated list of domain names whose profile × domain skills are merged.