land-and-deploy
Warn
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The skill is instructed to follow 'onboarding and consent directives' provided within
GSTACK_INSTRUCTION_BEGINblocks by thegstack-skill-starttool at load time. This allows the backend tool to dynamically alter the agent's instructions based on the environment or session status. - [DYNAMIC_EXECUTION]: Uses the
evalcommand to execute the output string produced by thegstack-diff-scopebinary, which dynamically calculates and sets shell variables based on repository state. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from multiple sources without strong boundary isolation.
- Ingestion points: Processes pull request bodies (sourced via
gh pr view) and onboarding blocks from shell command outputs directly as instructions. - Boundary markers: No explicit delimiters (like XML tags or triple backticks with 'ignore instructions' warnings) are used when piping external data like the PR body into subsequent shells or tools.
- Capability inventory: The skill has extensive permissions including the
Bashtool, the ability to push to remote git branches, merging PRs through CLI tools, and executing JS scripts in an external browser viaaside repl. - Sanitization: Uses a specialized tool called
gstack-issue-guardto scan the PR body for malicious content before processing, which serves as a mitigation step. - [COMMAND_EXECUTION]: Orchestrates complex release engineering tasks by executing a wide variety of shell commands, including Git operations (
git fetch,git push,git revert), GitHub CLI commands (gh pr merge,gh api), and various local gstack binaries located in~/.claude/skills/gstack/bin/. - [EXTERNAL_DOWNLOADS]: Integrates with the
asideagentic browser for canary verification. While the skill does not perform the download itself, it directs users toaside.comand relies on this external tool for critical verification steps, introducing a dependency on a non-standard browser environment.
Audit Metadata