The Agent Skills Directory

[COMMAND_EXECUTION]: The file scripts/create-integration-test.md contains multiple dynamic context injection patterns using the !command syntax. These patterns trigger automated shell execution of commands such as grep and find when the skill is processed, bypassing standard user review for tool execution.\n- [DATA_EXFILTRATION]: Within the dynamic context blocks in scripts/create-integration-test.md, the command grep -r "test_database\\|TEST_DB" .env* performs a recursive search across environment files. Accessing .env files is a high-risk operation as it can expose hardcoded credentials, API keys, and other secrets to the agent's context.\n- [PROMPT_INJECTION]: The skill provides explicit behavioral instructions in rules/verification-techniques.md under the 'Evidence Verification' section. These instructions mandate specific evidence collection protocols (e.g., exit codes, coverage targets) that override the agent's default logic for determining task completion.\n- [PROMPT_INJECTION]: An indirect prompt injection surface is identified as the skill is designed to ingest and process untrusted data from project files and external web sources. \n
Ingestion points: Uses Read, Glob, and Grep tools on local project files, and WebFetch/WebSearch for external content.\n
Boundary markers: The skill does not define specific delimiters or warnings to ignore embedded instructions in the processed data.\n
Capability inventory: The agent has access to file system tools, web search tools, and automated shell command execution.\n
Sanitization: No sanitization or validation logic is present to filter malicious instructions from the ingested content.

testing-integration