Action contexts (Slice 5a)

resolve_who_can_act answers "who can act on this task now?" by consulting the CONTEXT_REGISTRY against the live tool-status cache. Tasks declare agent_required_contexts + user_required_contexts; the resolver returns a WhoCanActDecision with per-side unmet tokens and a handoff-eligible flag.

Details

automation/contexts (Slice 5a)

Module: work_buddy/automation/contexts.py.

What

Two concepts:

1. CONTEXT_REGISTRY -- dict[str, list[str] | None] mapping context tokens (e.g. @filesystem, @email_send) to the tool IDs that satisfy them for the agent.
2. resolve_who_can_act(agent_required, user_required, *, tool_status=None) -- pure function returning a frozen WhoCanActDecision.

Sentinel discipline (registry values)

• None -> user-only context (@physical, @user_creds, @cluster). Agent never satisfies regardless of tool state.
• [] -> universally available (@filesystem, @web_public, @llm, @github). Both actors satisfy without a probe.
• [...] -> probe-gated. Agent satisfies iff ALL listed tool IDs are available (cf. work_buddy.tools.is_tool_available).

Starter token set (14)

User-only: @physical, @in_person, @phone_voice, @user_creds, @user_workstation, @cluster. Universal: @filesystem, @web_public, @llm, @github. Probe-gated: @vault -> obsidian, @email_send -> thunderbird, @email_read -> thunderbird, @chrome_active -> chrome_extension.

Unknown tokens (Clarify may invent new ones for forward-compat) are treated as user-only AND reported in unknown_tokens so the dashboard can warn.

WhoCanActDecision

Frozen dataclass: agent (bool), user (bool), blocked (bool), agent_unmet (tuple), user_unmet (tuple), agent_handoff_eligible (bool), unknown_tokens (tuple).

Handoff framing (ROADMAP section 3.2)

When agent=False AND user=True -> agent_handoff_eligible=True. The right surface is a HANDOFF card ("agent prepared what it can; you take from here"), NOT a "task waiting" badge. The Today tab renders this badge.

Risk integration

automation.risk.resolve_achievable_tier consults this module via the tool_status kwarg. When the agent cannot satisfy its required contexts, the achievable ceiling drops to 1 (suggest only). resolve_operating_tier then emits agent_context_unmet / user_context_unmet blockers per ROADMAP section 3.3.

Schema (task_metadata, Slice 5a)

Three nullable columns: - agent_required_contexts (TEXT, JSON array) - user_required_contexts (TEXT, JSON array) - required_contexts_source ('agent_inferred' | 'user_authored' | NULL)

Clarify populates the lists at task creation; the dashboard flips required_contexts_source to 'user_authored' once the user edits the inferred set.