The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a bash script in SKILL.md to iterate through directories and extract the description field from metadata files. The command cd /Users/engineer/workspace/backtest/.agents/skills; for d in */; do n="${d%/}"; desc=$(grep -m1 '^description:' "$n/SKILL.md" 2>/dev/null | sed 's/^description: //'); echo "$n :: $desc"; done is used to discover available tools in the local workspace.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection by ingesting data from external skill descriptions.
Ingestion points: The agent reads the description metadata from SKILL.md files located in subdirectories of the /Users/engineer/workspace/backtest/.agents/skills/ directory.
Boundary markers: None. The agent is instructed to read the list and categorize tools based on their descriptions without explicit instruction to ignore potential commands inside those descriptions.
Capability inventory: The agent can execute bash shell commands and generates a structured research plan that controls the subsequent steps of the research-market-workflow.
Sanitization: None. The content extracted from the external metadata files is processed directly by the model to decide the assembly of the 'research desk'.

research-manager