The Agent Skills Directory

[DYNAMIC_EXECUTION]: The diagnostic script scripts/smoke-ingest.sh uses a Python helper function that employs eval() to parse JSON responses from the logging API.
Evidence (scripts/smoke-ingest.sh): python3 -c 'import json,sys;d=json.load(sys.stdin);print(eval("d"+sys.argv[1]) if d else "")' "$1"
While the script currently uses hardcoded strings for the $1 parameter (e.g., '["ok"]'), the use of eval() to navigate data structures is a high-risk pattern that can be exploited for arbitrary code execution if the input is improperly handled.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to collect and process external application logs, which serves as an ingestion point for untrusted data that could influence downstream agents or analysis tools.
Ingestion points: Application logs collected from stdout, stderr, or local log files as described in SKILL.md and references/collectors.md.
Boundary markers: None explicitly enforced for the log message content itself, though the skill uses structured JSON for the transport layer.
Capability inventory: The skill provides tools for writing data to remote observability endpoints via curl and fluent-bit.
Sanitization: The skill provides clear guidance to developers to avoid logging sensitive identifiers such as tokens, passwords, and PII to reduce data exposure risks.
[COMMAND_EXECUTION]: The skill includes shell scripts and configurations that execute commands for log collection and network communication.
Evidence: Use of curl for API interaction in scripts/smoke-ingest.sh and fluent-bit for log forwarding in references/collectors.md.
These are legitimate functional requirements for the skill's purpose but represent a managed attack surface.

portal-logging