The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions require the agent to access highly sensitive project configuration files to discover the project's existing observability stack.\n
Evidence: Step 2 of the workflow in SKILL.md directs the agent to "Read docs, .env* names, SDK imports, dashboards, alert rules". Environment files (matching .env*) are high-sensitivity assets that commonly contain secrets, API keys, and private credentials. Accessing these files is a high-risk operation that could lead to accidental credential exposure.\n- [INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted data and possesses file-system modification capabilities, creating a potential vector for indirect prompt injection.\n
Ingestion points: The skill accepts user-provided arguments for "path|route|job|feature" to be instrumented.\n
Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands when the agent reads external documentation or project files.\n
Capability inventory: The agent can read project documentation and configuration and can write new instrumentation code (logs, traces, metrics) into the codebase.\n
Sanitization: While the skill includes high-level guardrails against logging PII and secrets, it does not specify technical validation or escaping for the user-provided strings used in the instrumentation process.\n- Remediation Guidance:\n
Instruct the agent to check for the presence of specific environment variable names rather than reading the entire content of .env files.\n
Require human review for all code modifications performed by the agent.\n
Ensure that all instrumentation data is properly sanitized before being passed to observability backends.

observe-it