The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands and scripts to manage the research corpus. Evidence: In SKILL.md, the agent is instructed to use 'bash scripts/init-corpus.sh' to initialize the directory structure. Evidence: In references/architecture/research-architecture.md, a Ruby script is used to verify markdown link integrity via the shell. Evidence: The skill uses find, wc, and rg for file management and verification as described in SKILL.md and references/architecture/research-architecture.md.
[EXTERNAL_DOWNLOADS]: The skill performs extensive web research by downloading content from various online sources. Evidence: The skill uses tools like web-search, scrape-links, and curl to gather data from vendor websites, documentation, Reddit, and Hacker News as detailed in references/agents/research-powerpack-and-explore.md.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from the web and maintains significant system capabilities. Ingestion points: Data enters the agent context through web search and scraping tools (web-search, scrape-links, curl) as referenced in references/agents/research-powerpack-and-explore.md. Boundary markers: The skill attempts to mitigate risks by using maximalist templates to structure output and requiring source maps and claims ledgers to track information provenance, as described in references/architecture/template-authoring.md and references/workflow/evidence-and-synthesis.md. Capability inventory: The agent can execute shell commands like bash, find, rg, mkdir, and curl across various files. Sanitization: The skill mandates the separation of facts from vendor claims and practitioner reports, and includes specific rules for attributing and labeling biased audience evidence (e.g., Reddit comments), as noted in SKILL.md and references/workflow/evidence-and-synthesis.md.

run-industry-research