The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the local codebase and external sources, presenting a vulnerability surface for indirect prompt injection.\n
Ingestion points: The skill reads arbitrary codebase files in SKILL.md (Step 1) and codebase-analyzer.md (Step 1). The external-research.md sub-agent further retrieves data from external URLs.\n
Boundary markers: There are no explicit delimiters or specific instructions for the agent to ignore or isolate instructions that may be embedded within the files it reads.\n
Capability inventory: The skill has access to tools for shell execution (Bash), file modification (Write), and integration with task management (linearis).\n
Sanitization: The skill lacks automated sanitization or filtering of external content before it is processed by the AI agents or written to documentation.\n- [EXTERNAL_DOWNLOADS]: The skill performs various network operations to fulfill its research and synchronization duties.\n
The external-research.md sub-agent uses the WebFetch and WebSearch tools to access arbitrary external resources.\n
SKILL.md invokes humanlayer thoughts sync and thoughts-pull-sync to fetch and update peer documentation state from a remote repository.\n- [COMMAND_EXECUTION]: The skill utilizes the Bash tool for host system interaction and metadata collection.\n
It executes git commands (rev-parse, branch) and proprietary CLI tools such as catalyst-session and linearis for project tracking.\n- [DYNAMIC_EXECUTION]: The skill dynamically spawns specialized sub-agents (codebase-locator, codebase-analyzer, etc.) by loading instructions from a set of provided markdown files within the skill's assets directory.

research-codebase