# Intent Check > Verify that the active primary's interpretation matches the operator's request before consequential planning or execution. Use when the user says "intent check," "confirm what I asked," "check for scope drift," or "/agent-collab:intent-check." Also offer this proactively when a major request has multiple constraints whose omission would materially change the result. - Skill: `sumitake/intent-check` (Agent Skill) - Install (CLI): `npx skillmds@latest add sumitake/intent-check` - Raw SKILL.md: https://api.skillmd.com/api/skills/sumitake/intent-check/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: sumitake (https://skillmd.com/u/sumitake) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/sumitake/intent-check --- ## Unified runtime invocation Resolve the **plugin root** from this loaded file: `SKILL.md` is at `/skills//SKILL.md`. Invoke only `python3 "/coordinator.py"` and send one bounded JSON routing request on EOF-delimited stdin, without a PTY. Use the Python invocation example in the **Routing request** section in `/README.md` and the co-packaged manifest's signed `wire_contract`; never invent fields or provider actions. Supply one caller-defined work unit per independently useful deliverable, with this skill's logical action and a bounded opaque payload. Use `depends_on` only for actual dependencies. Honor an operator-named provider with `explicit_target`. For an authorized independent review or governance task without an operator-named provider, also use that field to bind the caller-verified distinct reviewer selected by the caller or designated by the workflow. Carry the same target into planning and live dispatch; verify returned native lineage before accepting independence. Otherwise use normal untargeted routing. Choose quality and effort for the workload; include context/output token estimates when known. Read the current manifest digest and actual cwd device/inode; do not copy example values. The runtime owns its timeout; do not wrap it in a shorter fixed timeout. Repository identity, source-head verification, disposable copies, patch capture, and cleanup remain caller-owned where applicable. The shim runs standalone from the installed plugin and transports the routing client's bounded result without semantic interpretation. Never discover a provider executable, reconstruct a raw command, or replay, retry, or fail over a consumed work unit. Provider status, terminal records, receipts, telemetry, and other structured fields are optional diagnostics; none is a content-availability gate. Preserve every returned content record or recovered partial response and interpret it with ordinary model reasoning. Never synthesize approval, authority, or a receipt from process exit or missing diagnostics. A planning-only request sets `dispatch_requested=false`; a live request sets it true and consumes at most one provider attempt per work unit. Planning reports route eligibility, not live availability or authentication. Report a caller/client failure at that layer; provider state remains unknown unless native evidence establishes it. Content availability and each work unit's `execution_status` are separate facts. # Intent check - advisory interpretation comparison Freeze two artifacts: the operator's original request and the active primary's plain-language interpretation. Do not include an implementation plan; this step checks understanding, not design quality. Resolve the **plugin root** from this loaded file and invoke only `python3 "/coordinator.py"` with one normally untargeted `context.documents.intent` work unit. Put the verbatim operator request and plain-language primary interpretation in its bounded opaque payload. Use a provider-free planning request to inspect route eligibility before dispatch; set `explicit_target` only when the operator names a provider. The caller does not construct an agent/action pair or invent primary/author exclusion fields. This is an advisory context comparison and has no pre-dispatch family gate. ## Workflow 1. Quote the original request exactly, preserving negations and scope limits. 2. Write the interpretation as objective, in-scope work, out-of-scope work, constraints, success criteria, and stop conditions. 3. Put the two frozen documents in the payload and ask the reviewer to compare missed constraints, added scope, ambiguities, and a recommended interpretation. Do not add unsupported identity or family-exclusion fields. 4. Dispatch the advisory comparison and check that it addresses the frozen documents. Record the observed reviewer lineage and its source when available; otherwise label it lineage-unverified advisory. Same-family output remains advisory. Even with verified distinct lineage, document intent remains context only: it cannot satisfy a review or governance evidence contract. Preserve the complete raw response; do not require a verdict line or alternate envelope. 5. Adjudicate the returned text as match, drift, or ambiguity. On a match, proceed. On drift, revise the interpretation and recheck only when a new request is separately justified. On ambiguity, ask the operator only the load-bearing question. Preserve and reason over the full raw route result; never replay it for formatting. Never reconstruct a raw provider command or invoke a provider outside the co-packaged coordinator. Choose an explicit target only when the operator names its provider. A route-local failure does not establish that global governance is unavailable.