merge-pr
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub PR descriptions and review comments. This content is used to extract tasks and address feedback, which could serve as a vector for indirect prompt injection attacks designed to manipulate the agent's logic during the PR lifecycle.
- Ingestion points: PR descriptions are extracted in
references/post-merge.md(Step 12), and review comments are fetched via GraphQL inassets/references/merge-blocker-diagnosis.mdandassets/references/review-thread-resolution.md. - Boundary markers: The skill does not employ explicit boundary markers or instructions to ignore embedded commands when processing extracted prose from PRs.
- Capability inventory: The skill possesses high-privilege capabilities including git operations, PR merging, Linear ticket updates, and arbitrary shell command execution via test runners.
- Sanitization: Structured data is handled with
jq, but natural language content from PR descriptions is parsed usingsedandgrepwithout robust sanitization against adversarial instructions. - [COMMAND_EXECUTION]: In
references/pr-identification.md(Step 5), the skill executes a test command read directly from the.catalyst/config.jsonconfiguration file. While this is the primary mechanism for pre-merge verification, it represents a path for arbitrary command execution if an attacker can influence the repository's configuration. - [DATA_EXFILTRATION]: The skill reads Linear API tokens from local configuration files (
~/.config/catalyst/config-*.json) and sends them to the official Linear API (api.linear.app). This is the intended behavior for Linear integration but involves the transmission of sensitive credentials to an external service. - [EXTERNAL_DOWNLOADS]: The skill communicates with trusted and well-known services, including the GitHub GraphQL API and the Linear API. It also performs an HTTP smoke check on the vendor's domain (
catalyst.coalescelabs.ai). These operations are consistent with the documented functionality of the Catalyst platform.
Audit Metadata