The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements an evaluation harness that generates and executes JavaScript code artifacts. The script scripts/async-ownership-oracle.mjs uses dynamic import() to load and execute generated modules during experiments. Similarly, scripts/coverage-audit.mjs dynamically loads scenario files for analysis. These are core functional requirements for code evaluation and testing utilities.
[COMMAND_EXECUTION]: The skill spawns child processes to run its evaluation oracle and the agent harness. Specifically, scripts/async-ownership-pilot.mjs uses spawnSync and spawn to execute Node.js scripts and agent commands within a controlled workspace environment. The runner includes specific security flags (--no-session, --no-rules) to restrict the agent's capabilities during these tests.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of agent session logs and task definitions.
Ingestion points: scripts/session-extract.mjs reads JSONL session stores from user-provided directories; scripts/async-ownership-pilot.mjs reads TASK.md and GUIDANCE.md files as part of the experiment setup.
Boundary markers: Data is primarily handled as structured JSON or through regex-based parsing; explicit natural-language boundary markers (e.g., 'ignore previous instructions' guards) for the ingested data are not present in the utility scripts themselves.
Capability inventory: The skill possesses the capability to read/write files, execute shell commands, and dynamically load modules.
Sanitization: scripts/session-extract.mjs features a comprehensive redaction mechanism that uses regular expressions to filter sensitive information, including local home paths, email addresses, IP addresses, and hostnames, ensuring that analyzed session data is sanitized before aggregation.

optimising-skills