The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The setup routine performs remote script execution to install the Bun runtime.
Evidence: In the "SETUP" section of "SKILL.md", it downloads a script from "https://bun.sh/install". It implements a security check by comparing the file's SHA256 hash against a hardcoded constant ("bab8acfb046aac8c72407bdcce903957665d655d7acaa3e11c7c4616beae68dd") before execution.
[DYNAMIC_EXECUTION]: The skill uses shell evaluation to load configuration and project identifiers from its own local binaries.
Evidence: "SKILL.md" contains multiple instances of "eval "$(.../gstack-slug)"" and "source <(.../gstack-repo-mode)". These calls execute code generated by the skill's internal logic components to manage environment state.
[DATA_EXFILTRATION]: The skill manages telemetry and project artifact synchronization to external repositories.
Evidence: It calls "gstack-telemetry-log" and "gstack-brain-sync". These features are disclosed to the user through the "AskUserQuestion" tool, which explains the data shared (usage metadata or project plans/reports) and requires user selection of sync levels.
[INDIRECT_PROMPT_INJECTION]: The skill processes project metadata and collaborator-provided content which serves as a potential attack surface.
Ingestion points: Untrusted data enters via "gh pr view" (PR titles/bodies) and "git log" (commit messages) in "SKILL.md" (Steps 1 and 3.5).
Boundary markers: The skill instructs the agent to manually verify PR accuracy against actual code diffs, which acts as a logical check against misleading content.
Capability inventory: The skill has broad privileges including "Bash" command execution and "Write" access to project files during merge and revert operations.
Sanitization: No explicit technical sanitization is performed on the PR/commit text before it is used to inform the agent's context or generate the final report.

land-and-deploy