The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/delegate_to_rlm.py implements a discovery mechanism to find and execute the openplanter-agent from the local filesystem. It searches common developer paths (e.g., ~/Desktop/Programming/OpenPlanter, ~/projects/OpenPlanter) or follows the OPENPLANTER_REPO environment variable, executing the discovered code via subprocess.run.
[COMMAND_EXECUTION]: Several scripts, including scripts/investigate.py and scripts/web_enrich.py, use subprocess.run to orchestrate the investigation pipeline by calling sibling Python scripts and other Claude Code skills such as exa-search.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest, normalize, and analyze datasets from a wide variety of external APIs and bulk downloads. This ingestion of untrusted data from sources like the SEC, FEC, OSHA, and ICIJ creates an attack surface where malicious data could potentially influence the agent's reasoning during the analysis phase.
Ingestion points: scripts/dataset_fetcher.py, scripts/scrape_records.py, and specialized fetchers for SAM.gov, EPA, OSHA, ICIJ, Census, and ProPublica 990.
Boundary markers: The methodology in SKILL.md provides epistemic guidelines for the agent to verify ground truth from files, though the scripts do not implement strict LLM-level prompt delimiters for the data content.
Capability inventory: The skill can execute shell commands via subprocess, write to the filesystem, and make network requests.
Sanitization: scripts/entity_resolver.py performs normalization (Unicode NFKD, legal suffix stripping, etc.) on entity names before processing.
[EXTERNAL_DOWNLOADS]: The skill fetches bulk data and records from numerous well-known government and public record services including sec.gov, fec.gov, treasury.gov, lda.senate.gov, usaspending.gov, census.gov, epa.gov, icij.org, dol.gov, propublica.org, and sam.gov. These downloads are documented as safe functional requirements for the investigation methodology.
[DYNAMIC_EXECUTION]: The delegate_to_rlm.py script dynamically constructs and executes commands to spawn a full RLM agent, including passing user-defined objectives and API keys for various LLM providers.

openplanter