The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses potentially destructive Git commands, specifically git add -A and git reset --hard, to implement the 'Revert' primitive in the TCR cycle. Evidence is found in references/git-cycle.md and SKILL.md. The skill provides clear warnings that these commands will delete untracked files without recovery.
[COMMAND_EXECUTION]: The skill executes arbitrary shell commands defined as <test_command>. It attempts to discover these commands from project metadata like package.json, Makefile, or language-specific conventions (e.g., pytest, cargo test).
[DYNAMIC_EXECUTION]: The workflow involves building and executing shell commands at runtime based on environment discovery. For example, it concatenates git add -A with the detected test command and executes the result to determine the workflow phase outcome.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes the output of external test runners to decide whether to commit or revert code. An attacker who can influence the output of a test could potentially influence the agent's decision-making process.
Ingestion points: Test runner stdout/stderr and exit codes (referenced in references/git-cycle.md).
Boundary markers: None identified for external test output.
Capability inventory: File system modification via git (add, commit, reset) and arbitrary test command execution.
Sanitization: No specific sanitization or filtering of test output is documented before the agent evaluates it.

tcrdd