tally-rewind
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The
scripts/measure.shscript downloads thephpstan/tallytool usingcomposer. This is an official tool provided by the skill author to enable the benchmarking functionality. - [COMMAND_EXECUTION]: The skill utilizes several CLI tools including
gitfor history traversal,composerfor dependency management, andghfor interacting with GitHub. It also allows users to provide an optional setup command via theTALLY_REWIND_PREPAREenvironment variable to configure the project environment before benchmarking. - [DYNAMIC_EXECUTION]: The benchmarking script dynamically generates a PHP configuration file (
.tally-rewind.php) at runtime to configure the Tally execution environment and project autoloaders. - [INDIRECT_PROMPT_INJECTION]: The skill analyzes potentially untrusted data from git commit messages, changelogs, and pull request descriptions to identify historical performance improvements.
- Ingestion points: Commit messages (
git log), pull request metadata (gh), and project changelog files. - Boundary markers: The agent is instructed to parse these for specific keywords and semantic meaning related to performance, rather than executing them as instructions.
- Capability inventory: The skill possesses capabilities to execute shell commands and PHP code through its benchmarking scripts.
- Sanitization: No specific sanitization is performed on the ingested text beyond the LLM's natural language processing.
Audit Metadata