The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to interact with git, gh, and glab CLIs. It performs operations like git rebase, git push --force-with-lease, and directory creation. While the instructions include safety checks (preventing execution on protected branches and requiring clean working trees), these are powerful tools that modify repository state.
[INDIRECT_PROMPT_INJECTION]: The skill parses untrusted data from git log (commit subjects and bodies) and interpolates them into markdown tables and prompt contexts for review. An attacker could craft a commit message containing instructions that attempt to influence the agent's behavior during the 'retcon' or 'preview' workflows. The skill lacks explicit boundary markers or sanitization for this external content.
[DYNAMIC_EXECUTION]: In the retcon.md workflow, the skill dynamically generates shell scripts (TMPSCRIPT, MSG_SCRIPT) using mktemp and cat to serve as GIT_SEQUENCE_EDITOR and GIT_EDITOR during an interactive rebase. This technique is used to automate commit message rewriting but involves the runtime creation and execution of scripts based on parsed commit data.

changelog