The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses eval and source to execute commands generated by external binaries located in the user's home directory. This allows for arbitrary command execution if those binaries are compromised or return unexpected content.
Evidence in SKILL.md (Preamble): eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)"
Evidence in SKILL.md (Preamble): source <(~/.claude/skills/gstack/bin/gstack-repo-mode 2>/dev/null)
Evidence in SKILL.md (Context Recovery): eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)"
[COMMAND_EXECUTION]: The skill automatically detects and executes test and linting commands found in the project's configuration files (e.g., package.json, pyproject.toml). It also dynamically parses package.json to extract and run shell commands.
Evidence in SKILL.md (Step 1): TEST: $(node -e "console.log(JSON.parse(require('fs').readFileSync('package.json','utf8')).scripts.test)" 2>/dev/null)
Evidence in SKILL.md (Step 2): tsc --noEmit 2>&1 | tail -50 (and other detected tools)
[INDIRECT_PROMPT_INJECTION]: The skill ingests and displays output from external developer tools (compilers, linters, test runners). While the skill includes instructions for the agent to not act on these findings automatically, the processing of potentially attacker-controlled output from these tools presents a vulnerability surface.
Ingestion points: Standard output and error streams from tsc, biome, eslint, ruff, pytest, cargo test, knip, and shellcheck (SKILL.md, Step 2).
Boundary markers: The skill uses a HARD GATE instruction stating 'Do NOT fix any issues. Produce the dashboard and recommendations only.' (SKILL.md, invocation instructions).
Capability inventory: The skill has access to Bash, Write, Edit, and AskUserQuestion tools.
Sanitization: The skill captures the last 50 lines of raw output for reporting without explicit sanitization or escaping (SKILL.md, Step 2).
[EXTERNAL_DOWNLOADS]: The skill includes an 'Artifacts Sync' feature that synchronizes project artifacts (plans, designs, reports) to a remote GitHub repository. While this is an opt-in feature, it involves network operations and data transfer to external infrastructure.
Evidence in SKILL.md (Artifacts Sync): References to gbrain search, gbrain query, and gstack-brain-sync binaries that handle remote synchronization.

health