The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes potentially untrusted data from subagent reports, logs, and screenshots, creating a vulnerability to indirect prompt injection.
Ingestion points: SKILL.md instructs the agent to ingest and analyze "subagent reports", "giant logs", and "query dumps".
Boundary markers: The instructions lack requirements for using protective delimiters or "ignore instructions" tags when reading external data.
Capability inventory: The skill uses the ax CLI tool and dispatches subagents that can interact with the filesystem and execute tasks.
Sanitization: No sanitization or validation logic is defined for the content of ingested external files.
[COMMAND_EXECUTION]: The skill utilizes the ax CLI utility to perform cost analysis and verify routing (e.g., ax dispatches, ax cost split, ax cost images). It accesses local configuration and hook files located at ~/.ax/hooks/ and references the @ax/hooks-sdk package.

efficient-dispatch