The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The resources/error-playbook.md file contains instructions for the agent to install the curl_cffi Python package using pip if the impersonation strategy fails. While this uses a well-known package registry, runtime dependency installation is a security concern.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external web sources and remote code repositories, presenting a risk of indirect prompt injection.
Ingestion points: Data is fetched from external URLs via oma search fetch and from GitHub/GitLab repositories using gh and glab commands, as described in resources/execution-protocol.md.
Boundary markers: The skill does not implement explicit boundary markers or delimiters to separate untrusted external content from agent instructions.
Capability inventory: The skill possesses significant capabilities, including executing CLI tools (gh, glab, oma), spawning browser processes, and interacting with local code intelligence providers.
Sanitization: Although the skill normalizes and deduplicates results, it does not explicitly sanitize or filter the content for embedded malicious instructions.
[DYNAMIC_EXECUTION]: The web retrieval process involves dynamic execution strategies, such as using puppeteer-core for browser-based fetching and spawning subprocesses via curl_cffi to impersonate different browsers.
[COMMAND_EXECUTION]: The skill regularly spawns subprocesses to execute external CLI tools like gh (GitHub CLI), glab (GitLab CLI), and various oma search primitives, passing user-provided queries as arguments.

oma-search