States completion evidence required before a change is mergeable—automated checks, manual review triggers, artifact retention, and explicit waivers—so AI-generated work cannot skip verification silently.
Package status: AI-assisted reference package. Static package validation has passed; implementation scenarios remain not run and human domain review is required before stable adoption.
Decision
Define done as mergeable evidence, not narrative confidence. Every change class lists required automated checks, manual review triggers, and retained artifacts. AI-generated contributions follow the same gates plus explicit disclosure of tests run, skipped, and why.
Scope
- Change classification and per-class verification requirements.
- Merge gate behavior, waiver rules, and evidence retention.
- Additional requirements when an AI agent authors or materially edits the change.
- Linkage between acceptance scenarios and verification results.
Outside this block
- Detailed test strategy for individual modules (see unit testing block).
- Protected path policy (see protected-files block).
- Production deployment approval separate from merge.
Contract
- Each pull request declares a change class; undeclared classes default to CLS-FEATURE as the strictest applicable profile default.
- Required checks for the declared class must pass or carry an approved, time-bounded waiver with owner.
- Manual review triggers map to reviewer roles in
example.yaml; touches-authentication requires security-reviewer. - AI-authored changes include a human-readable diff summary and an explicit list of tests not executed with rationale.
- Linked acceptance scenario IDs must appear in the change description when behavior changes; missing links block merge for feature classes.
- Verification evidence remains queryable for the profile retention period after merge.
Implementation guidance
- Implement merge gates in CI using the change class matrix from
example.yaml. - Store check results and waiver tickets in durable audit storage linked to pull request and commit SHA.
- Train AI agents to emit the required summary template before requesting review.
- Align acceptance scenario IDs with PureIntent blocks or project-local equivalents stored in the same repository.
Failure handling and safeguards
- Waivers cannot bypass protected-path rules or security review triggers.
- Post-merge follow-up jobs reopen the pull request or file defects if waived checks remain failing after the profile window.
- Flaky checks are quarantined with owner and expiry; they cannot be permanently waived.
Verification and operations
- Measure waiver rate, mean time to green checks, and AI change rework rate.
- Audit monthly that change-class rules still match repository layout.
- Sample merged AI changes for undisclosed skipped tests.
The executable-looking examples in this package are fixtures and acceptance contracts. Validate package structure before adoption; then implement and execute the scenarios in the target repository.
Adoption assumptions
- Check names, retention days, and waiver duration in
example.yamlare profile choices, not universal defaults. - Teams must map triggers to their reviewers and ticketing system.
References
- No external normative source is required; this package defines merge verification policy.