Run one adversarial robustness round over a codebase - find defects by operating and asserting, fix each with a guard in the same commit, mutation-verify every guard, and write the round record. Use when asked to harden, audit, stress, or run a robustness round on a repo.
Installs into .claude/skills of the current project.
Are you the author of Robustness Loop?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/sutradharhq-robustness-loop)
---
description: Run one adversarial robustness round over a codebase - find defects by operating and asserting, fix each with a guard in the same commit, mutation-verify every guard, and write the round record. Use when asked to harden, audit, stress, or run a robustness round on a repo.
---
# Robustness loop
The procedure is kept in one canonical file so that every harness reads the
same text. Read it now, in full, and follow it:
${CLAUDE_PLUGIN_ROOT}/skills/robustness-loop/robustness-loop.md
That file is a byte-identical copy of `agent/skills/robustness-loop.md` in
<https://github.com/sutradharhq/sutradhar>, kept in step by
`plugin/sync_guards.py`. If it is not there, this copy of the plugin is
incomplete. Say so rather than improvising the procedure from memory - the
value of the loop is in its specifics, and a half-remembered version of it
is a code review with a longer name.
The guards the loop calls for are available as MCP tools from the
`sutradhar-guards` server this plugin registers, and as CLIs under
`python/sutradhar_guards/`.
