The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external workspace files which may contain untrusted content. If these files contain malicious instructions, they could influence the agent's behavior when processing the generated table.
Ingestion points: Reads literature taxonomy from workspace/outline/taxonomy.yml and paper notes from workspace/papers/paper_notes.jsonl in scripts/run.py.
Boundary markers: The script does not implement explicit delimiters or warnings to the agent to ignore instructions embedded within the research notes.
Capability inventory: The script has the capability to write files to the workspace via write_markdown and write_jsonl.
Sanitization: There is no evidence of sanitization or filtering of the content read from the YAML or JSONL files before it is processed and written to the output table.
[DYNAMIC_EXECUTION]: The script uses dynamic path resolution to modify the module search path at runtime to load local dependencies.
Evidence: In scripts/run.py, the script iterates up the directory tree to find a root directory containing AGENTS.md and then uses sys.path.insert(0, str(repo_root)) to allow importing the tooling module. While common in local development tools, this represents dynamic loading from a computed path.

idea-signal-mapper