The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands to validate and move files. Specifically, it instructs the use of dagr check to validate candidate JSON files and mv to atomically publish them. It also suggests falling back to cargo run if the binary is not found in the path.
[PROMPT_INJECTION]: The skill defines a mechanism for 'Indirect Prompt Injection' by processing external data via the [DAGR OPERATOR MESSAGE] envelope. This represents a surface where untrusted data (operator prose) enters the agent's context and is treated as instruction.
Ingestion points: External prose is ingested through the [DAGR OPERATOR MESSAGE] envelope, as described in the 'Handle an operator message' section.
Boundary markers: The skill uses a specific text envelope [DAGR OPERATOR MESSAGE] to delimit these instructions and explicitly warns the agent to treat the prose as bounded by the existing run scope.
Capability inventory: The agent has the capability to write to the local filesystem (run.json), execute the dagr validation binary, and perform atomic renames (mv).
Sanitization: The skill relies on the dagr check --strict --json command to ensure that any changes made to the run file conform to the required JSON schema before publication.

dagr-producer