The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local project files to generate a downstream configuration file.
Ingestion points: Reads GOAL.md and DECISIONS.md to identify source candidates as described in SKILL.md workflow.
Boundary markers: The skill does not use explicit boundary markers or delimiters when processing these inputs.
Capability inventory: The script scripts/run.py can write to the filesystem (sources/manifest.yml) using atomic_write_text.
Sanitization: The script performs structural validation of the YAML and checks for placeholder tokens, but does not sanitize URL or file path content against malicious payloads.
[DYNAMIC_EXECUTION]: The script dynamically modifies the system search path to load internal modules.
Evidence: scripts/run.py uses sys.path.insert(0, str(repo_root)) after resolving the repository root by searching upward for an AGENTS.md file. This allows it to import shared tooling.common utilities.

source-manifest