The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads external repositories using git clone in references/repo.md and fetches remote web content via WebFetch in agents/dw-researcher.md to verify documentation and references.
[DATA_EXFILTRATION]: The skill accesses sensitive local directories containing session history and transcripts from other AI platforms, including ~/.claude/projects/, ~/.pi/agent/sessions/, ~/.codex/sessions/, ~/.local/share/goose/sessions/, and ~/.local/share/opencode/storage/session/. It further instructs the agent to search for and access additional transcript locations if the specific tool is not listed.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute shell commands such as git clone, diffwiki CLI operations, and file system management through mktemp and rm commands.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Reads untrusted content from local session history files, external codebases (READMEs, manifests, ADRs), and remote web pages (references/history.md, references/repo.md, agents/dw-researcher.md).
Boundary markers: The agents/dw-author.md and agents/dw-researcher.md agents include instructions to treat captured text strictly as data and not as new instructions.
Capability inventory: The skill possesses Bash execution, network fetching (WebFetch), and comprehensive file system access (Read, Grep, Glob).
Sanitization: Implements a mandatory redaction rule in SKILL.md and references/repo.md to prevent secrets or tokens found in sessions or code from being recorded in the generated wiki entries.

dw-learn