The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several command-line tools to perform maintenance and security tasks.
Evidence: Uses git for commit, push, and history analysis in GitPush.md and DocumentRecent.md.
Evidence: Executes trufflehog via the SecretScan.ts utility to detect exposed credentials.
Evidence: Runs local TypeScript tools using bun run for index management and search operations.
[PROMPT_INJECTION]: Implements a customization system that allows overriding default skill behavior through local files.
Evidence: SKILL.md directs the agent to load and apply PREFERENCES.md and other resources from ~/.opencode/skills/PAI/USER/SKILLCUSTOMIZATIONS/System/ if the directory exists.
[PROMPT_INJECTION]: Maintains an indirect prompt injection surface by ingesting and processing session transcripts and filesystem content.
Ingestion points: Reads transcript history in DocumentSession.md and audits the ~/.opencode/ directory structure in IntegrityCheck.md.
Boundary markers: No explicit delimiters or "ignore embedded instructions" warnings are specified for the ingested content during processing.
Capability inventory: Capabilities include file writing (Bun.write), subprocess execution (git, trufflehog, curl), and spawning sub-agents via the Task tool.
Sanitization: Employs redaction for detected secrets in SecretScan.ts to prevent logging raw credentials.
[DATA_EXFILTRATION]: Performs network and git operations to manage system notifications and source code.
Evidence: Sends POST requests to http://localhost:8888/notify for local voice notifications.
Evidence: Pushes code to the author's repositories (Steffen025/jeremy-2.0-claudecode and Steffen025/pai-opencode) using the GitPush.md workflow.
[EXTERNAL_DOWNLOADS]: References the installation of external security tools from well-known sources.
Evidence: SecretScanning.md and SecretScan.ts recommend installing the trufflehog security scanner via Homebrew.

System