The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python script (scripts/peer_impact.py) and a test shell script (tests/test_peer_impact.sh). The Python script uses standard libraries for data processing (argparse, json, sys). The shell script is for testing purposes, resolving its own path and executing the Python script with various arguments to verify ranking logic and margin arithmetic. All command execution is local, deterministic, and scoped to the skill's primary analytical purpose.- [DATA_EXPOSURE]: The skill processes financial segment data and competitor lists from a local JSON file (tests/fixture_peers.json). It does not access sensitive system paths, environment variables, or user credentials. The logic focuses on publicly available financial filing data.- [REMOTE_CODE_EXECUTION]: There are no patterns of remote code execution, package installations, or external script fetching. All dependencies are from the Python and Bash standard libraries.- [PROMPT_INJECTION]: The instructions focus strictly on financial research methodologies and output formatting. There are no attempts to override agent safety filters, extract system prompts, or adopt unrestricted personas.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a JSON candidate list (--candidates). However, this data is strictly processed by a Python script for ranking and arithmetic rather than being interpolated into an LLM prompt for autonomous action. The script includes boundary checks (e.g., segment share totals) and warns about missing evidence, providing a layer of validation for external data.

peer-impact