| name | context-status |
| description | Read side of the consigliere workspace: answer a manager's natural-language question ('how's payments?', 'what's blocked on checkout-revamp?', 'who owns X?') from the materialized state — grounded, read-only, with source citations and the consigliere lens. Says 'not recorded' rather than inventing; never writes. Manual; consigliere bundle.
|
| triggers | {"keywords":["context status","how's","what's blocked","status of","consult my workspace","what do we know about"]} |
Context Status
Overview
digest-source fills the workspace; context-status reads it. Two weeks
after a meeting the manager asks "how's payments? what's blocked?" and wants a
grounded answer assembled from the materialized state — not a re-read of every
transcript. This skill answers natural-language questions over the workspace,
read-only and strictly grounded. It is the consult half; capture is
digest-source, and the heuristics loop is playbook-review.
When to Engage
Manual, operator-run. Engage when the manager asks about the current state of a
context or project — status, blockers, owners, decisions, the system map, or the
political risk. Not auto-invoked.
Invocation
/octopus:context-status "<natural-language question>"
# or a path: /octopus:context-status payments
Step 1 — Resolve workspace (read-only)
Read the consigliere.workspace config key. If it is unset, refuse and point the
manager to /octopus:consigliere-bootstrap. All reads stay within the resolved
workspace — this skill reuses the consigliere-bootstrap write-guard as a read
boundary and never writes anything.
Step 2 — Interpret the question + route
Map the question to a target — a context path or a project — from the question text
and the existing contexts/ tree and projects/*/meta.yml. When the target is
ambiguous, show what you resolved and let the manager correct it. Ambiguity → ask,
never guess (the same routing discipline as digest-source).
Step 3 — Read the materialized state
Read the target node's state.md — the six fixed sections (Status by workstream,
Blockers, Decisions, System & area map, Actions, Political risk). A context's
state.md carries fan-out pointer lines to the projects it touches; follow a pointer
into a project's state.md only when the question needs the detail. Read
journal.md only when the question is about history ("when did X get blocked?").
Read no more than the question requires.
Step 4 — Answer, grounded, with the consigliere lens
Answer concisely, leading with exactly what was asked (status, blocker, owner, risk).
Every claim carries its provenance — the state.md section or the (src: …) anchor
behind it. Apply the consigliere role's lens: surface a relevant political
risk or an approved heuristic ("owner tends to delay → consider a FUP") as a
grounded note, never as invented fact.
Strict grounding (reuses audit-grounding): answer only from the recorded state,
journal, and sources. If the answer is not recorded, say so plainly and suggest
running /octopus:digest-source on the relevant input — do not guess, and do
not fill the gap from outside the workspace.
Anti-patterns
- Answering beyond the recorded state — say "not recorded" instead of inventing.
- Writing anything — this skill is strictly read-only.
- Guessing the target on an ambiguous question instead of asking.
- Reading every file when the question only needs one
state.md.
Related
- Depends on
consigliere-bootstrap (the workspace + write-guard) and the
state digest-source writes.
- Reuses
audit-grounding and applies the consigliere role.
- Sibling:
playbook-review owns the heuristics loop this skill applies from.
