The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture requires the agent to read an external instruction file (engine.skillMd) at runtime and treat it as the authoritative protocol for the research task.\n
Ingestion points: Instructions are ingested from the engine.skillMd file located in the dynamically resolved engine root directory.\n
Boundary markers: The skill lacks boundary markers; it explicitly directs the agent to follow the external file's instructions exactly and top to bottom.\n
Capability inventory: The skill possesses the ability to execute shell commands via oma market run and write research briefs to the filesystem.\n
Sanitization: There is no evidence of sanitization performed on the content of the external instruction file before it is adopted into the agent's context.\n- [EXTERNAL_DOWNLOADS]: The skill uses the oma toolset to synchronize and update the research engine from GitHub.\n
The oma market update and oma market resolve commands fetch engine components from a remote GitHub repository.\n
These resources are managed by the vendor (first-fluke) as part of the skill's standard research functionality.\n- [COMMAND_EXECUTION]: The skill executes external CLI tools and scripts to perform research and validation.\n
It runs the oma market run command to execute Python-based research logic.\n
It utilizes oma market detect-trap as a preflight gate to validate research topics against safety rules.\n
The agent is instructed to perform self-checks by executing shell commands (grep, head) against generated research briefs.\n- [DYNAMIC_EXECUTION]: The skill's control flow is determined by environment resolution at runtime.\n
The oma market resolve command identifies script and configuration paths that the agent then loads and follows.\n
The agent dynamically adopts logic from the resolved engine.skillMd file to guide its multi-step research process.

oma-market