Context

Unified runtime invocation

Resolve the plugin root from this loaded file: SKILL.md is at <plugin-root>/skills/<skill-name>/SKILL.md. Invoke only python3 "<plugin-root>/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 <plugin-root>/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.

Source-grounded context

Use the public context capability for read-only summarization, extraction, comparison, synthesis, dependency/control-flow tracing, repository inventory, log or transcript audit, and anomaly or gap identification. It is not a governance verdict even when the prompt says “review” or “audit.”

Choose exactly one source mode

• Documents: provide bounded UTF-8 documents objects containing only label and content. Select context.documents.extract for mechanical structured extraction or context.documents.reason for multi-source synthesis and reasoning.
• Repository: provide the canonical repo_root and exact expected_repo_head. Select context.repository.extract for inventory or tracing, or context.repository.reason for architectural or multi-hop synthesis.

Reject prompt-only requests, hybrid documents-plus-repository requests, document paths/globs/file handles, unextracted binary input, and source escapes. Repository mode runs from the sealed repository root; document mode runs from an empty request-private document root.

Request and result contract

Resolve the plugin root from this loaded file, read the routing schema in <plugin-root>/README.md, and submit one work unit to python3 "<plugin-root>/coordinator.py". Use an explicit_target only when the user named one; never construct a provider transport action. Use quality_profile="economical" with effort_class="minimal" for mechanical extraction, and raise these closed provider-neutral fields only when the task actually requires more synthesis depth. Never name a model or provider member.

One accepted request launches one provider process and fresh session. Provider- internal tool rounds or model calls may exceed one. There is no automatic whole-request replay after any model call and no malformed-output retry. Preserve typed failures.

Preserve every bounded returned content frame or recovered partial. Interpret the raw text with ordinary reasoning, then spot-check load-bearing claims against the caller-owned document or exact repository source. Missing or conflicting diagnostics do not hide content and do not create authority.

This capability is advisory and read-only. It never edits files, applies a patch, creates governance evidence, or authorizes a provider command outside the co-packaged runtime.