Gate Runner
| This documentation was generated with the assistance of AI. Please report any inaccuracies. |
A custom agent that runs a single verification/quality-gate skill — tests, coverage, code-quality/lint, license headers, doc-comment audits, security scans, or a full project build — in an isolated context, optionally diffs the result against a baseline, and reports back only a compact, structured summary.
Purpose
iru-gate-runner exists to collapse a pattern that used to be hand-written, nearly identically, at more than a
dozen call sites across iru-code, iru-code-one-task-group, iru-java-code-one-task-group, and
iru-dotnet-code-one-task-group: invoke a verification skill,
diff it against a baseline if one is given, and return only the distilled outcome — never the raw
Surefire/JaCoCo/Checkstyle/PMD/SpotBugs/StyleCop/CA/SARIF/detect-secrets output or a full build log. Before this
agent existed, every one of those call sites repeated the same "report back only X, not the full report"
instruction in its own Agent() prompt; centralizing it in one agent definition means that convention only has
to be stated — and changed, if it ever needs to be — once.
It always prefers invoking the matching Claude Code skill (e.g. iru-java-test, iru-dotnet-coverage,
iru-java-code-quality, iru-check-license, iru-check-security) over a raw command, falling back to the equivalent raw
build-tool command only when the prompt says no matching skill is installed.
Tools
Skill, Bash, Read — enough to invoke a skill (or fall back to a raw command) and read whatever report file
it produces; no Edit/Write access, since this agent only runs and reports, it never modifies files itself.
Used by
-
Code — Step 3 (project-wide quality and security baseline capture) and Step 7 (full local build verification, plus the final quality/security baseline comparison).
-
Code One Task Group — the best-effort fallback path (Step 3), for a task whose language/framework has no matching
<key>-code-one-task-groupskill installed. -
Java Code One Task Group and .NET Code One Task Group — every delegated check while validating one task group’s bucket: pre-change quality baseline, license headers, Javadoc/XML doc comments, scoped test run, coverage check, full test suite, and the post-change quality diff — all run once per bucket rather than once per task.
-
Plan — recommended (when installed) for the scoped test-run step a generated
implementation_plan.mdinstructs its executor to delegate to a sub-agent, in place of a generic, repository-specific sub-agent prompt.
Why this one specifically
Every one of these delegations exists to keep something noisy (a report, a build log) out of the calling
skill’s own context window, and to diff against a baseline so only newly-introduced issues are ever reported —
not to use any capability unique to a broadly-scoped agent. A narrow, purpose-built agent with Skill/Bash/
Read-only tool access is a tighter fit than the default general-purpose type, and lets every caller state the
delegation in one line ("run X, diff against this baseline, report Y") instead of re-explaining the
summarize-don’t-dump convention every time.
Source
gate-runner.md on GitHub — the file this page was generated from.