The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and processes session transcripts (.jsonl files) to generate learning reports.
Ingestion points: The skill reads files from $HOME/.claude/projects/*.jsonl (SKILL.md, Step S1).
Boundary markers: No explicit boundary markers are used when reading input transcripts, though output is structured via markdown templates.
Capability inventory: The skill uses the Write tool to create files in $WK_SKILLS_HOME/learnings/ and Bash utilities (find, jq, sed) to process logs.
Sanitization: The skill implements a robust "HARD RULE" (Step 3) requiring the scrubbing of internal project names, file paths, secrets, employer/org names, PR/issue numbers, and human usernames before any data is written.
[COMMAND_EXECUTION]: The skill uses shell commands to automate the discovery and parsing of log files.
Evidence: Uses find to locate .jsonl files and jq to extract message content from the JSON structures. These operations are scoped to the user's project transcript directory and use static filter strings.

wk-learn