The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill makes extensive use of eval and source on the output of local binaries in its preamble and context recovery sections.\n
Evidence: eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)" is executed twice.\n
Evidence: source <(~/.claude/skills/gstack/bin/gstack-repo-mode 2>/dev/null) is used to dynamically load environment variables or functions.\n- [PERSISTENCE]: The skill includes logic to automatically modify and commit changes to the project's CLAUDE.md file to inject 'Skill routing' instructions.\n
Evidence: The skill appends a section to CLAUDE.md and runs git commit to persist agent behavior modifications across the repository.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted local data files and project configuration which could serve as a vector for indirect prompt injection.\n
Ingestion points: Reads from ~/.gstack/projects/{repo_slug}/learnings.jsonl, ~/.gstack/analytics/eureka.jsonl, and CLAUDE.md.\n
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are used for these data sources.\n
Capability inventory: The skill has access to shell execution (Bash), file writing (Write/Edit), and network operations via tools and its own binaries.\n
Sanitization: No sanitization is performed on the content of these ingested files before they are processed by the agent.\n- [DATA_EXFILTRATION]: The skill includes built-in telemetry and artifact synchronization features that transmit local data to external sources.\n
Evidence: gstack-telemetry-log transmits skill usage, duration, and repository name.\n
Evidence: gstack-brain-sync synchronizes artifacts (including plans and designs) to a private GitHub repository.\n
Note: These features are presented to the user for consent via AskUserQuestion before activation.

investigate