Add agentic workflow rule and verification eval#25
Merged
Conversation
Adds a global starter rule covering how Claude should operate during agentic work — not just what good code looks like: verify before claiming done, plan large changes, scope discipline, match surrounding code, and confirm irreversible actions. Pairs the rule with a workflow-verification eval (4 tests) so the standard is behaviorally checked, in keeping with staghorn's eval-driven approach. - internal/starter/rules/agentic-workflow.md: new global rule - internal/starter/evals/workflow-verification.yaml: matching eval - rules_test.go: register the rule in expectation lists - README.md: 25 -> 26 evals, add Workflow category - CHANGELOG.md: entry under [Unreleased]
HartBrook
added a commit
that referenced
this pull request
Jun 8, 2026
* Add verify starter skill Adds a runnable skill that operationalizes the agentic-workflow rule's "verify before claiming done": it discovers the project's own build and test commands, runs them, observes the actual output, and reports the result honestly rather than asserting success from inspection. Unlike the read-only starter skills, verify includes Bash so it can actually run the checks. Skips checks the project doesn't have instead of inventing commands. - internal/starter/skills/verify/SKILL.md: new skill - skills_test.go: register verify in expectation list - README.md: 3 -> 4 starter skills - CHANGELOG.md: entry under [Unreleased] * Stamp v0.9.0 release in changelog Move the unreleased entries (agentic-workflow rule + verification eval from #25, verify skill from #27) into a dated 0.9.0 section and update the compare links. Minor bump: all changes are new backward-compatible starter content. The v0.9.0 tag pushed after merge triggers the goreleaser release. * Update CHANGELOG.md Co-authored-by: Will Pike <6687499+pike00@users.noreply.github.com> * verify skill: also consult project docs (README, CONTRIBUTING, TESTING) for commands A project's canonical build/test commands are often documented in prose, not just in config files. Direct the skill to read those and treat a documented command as authoritative. --------- Co-authored-by: Will Pike <6687499+pike00@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The starter framework described what good code looks like but said almost nothing about how Claude should operate during agentic work. This adds that missing declarative layer.
agentic-workflow.md— a new global starter rule (applies to all files): verify before claiming done, plan large changes, scope discipline, match surrounding code, and confirm irreversible/outward-facing actions.workflow-verification.yaml— a matching eval (4 tests) so the standard is behaviorally checked, not just documented — in keeping with staghorn's eval-driven approach.Changes
internal/starter/rules/agentic-workflow.md— new ruleinternal/starter/evals/workflow-verification.yaml— new eval: recommends verification, scope discipline, surfacing tangents, honest reportinginternal/starter/rules_test.go— register the rule in both expectation listsREADME.md— 25 → 26 starter evals, new Workflow category rowCHANGELOG.md— entry under[Unreleased]Testing
go build ./...cleango test ./...passesLoadStarterEvalsloader (parses, 4 tests, validation passes)Follow-ups (intentionally out of scope)
verifyskill that runs the workflow (build/tests → honest report) as a counterpart to the rule.settings.jsonhooks/permissions bootstrap. Deterministic enforcement (format-on-stop, block-secret-commit, least-privilege) would be a new capability and warrants its own PR.