The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a series of local CLI subcommands (oma market detect-trap, harvest, score, fuse, cluster, render) to perform data processing. This deterministic approach reduces the LLM's direct exposure to raw external data during the high-risk acquisition phase.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It ingests untrusted content from community sources like Reddit, HN, and GitHub Issues. This content is processed by the LLM in the 'Analyst Synthesis' phase to generate SWOT and Porter's 5 Forces frameworks.
Ingestion points: Data enters the system via oma market harvest from third-party community platforms.
Boundary markers: The CLI pipeline aggregates data into a structured 'Cluster Bank' before LLM processing, providing a logical boundary between raw data and the synthesis prompt.
Capability inventory: The LLM's primary capability in this skill is writing the final report to the .agents/results/market/ directory.
Sanitization: The output-laws.md file defines strict rules for the final output, such as banning em-dashes and enforcing specific header structures, which acts as a form of output sanitization.
[EXTERNAL_DOWNLOADS]: The skill performs network operations indirectly by calling oma-search to fetch data from well-known community services (Reddit, Hacker News, Mastodon, etc.). It does not perform direct HTTP requests or execute remote scripts. Access to paid sources is guarded by environment variables handled at the CLI level.

oma-market