fix: order issue-type keyword rules by signal strength

[eleshar]

Linked issues

Fixes #3577

Found by Qodo review of #3573, which surfaced it against code that predates that fix.

Context

• Severity/Impact: Medium. Wrong issue types on project fields; no error, and the misclassification is invisible until someone reads the board.
• Affected versions/environments: scripts/agents/includes/derive-project-fields.cjs on develop. Pre-existing, not a regression from fix: correct stale issue-type inference and CommonJS title normalisation #3568.

Reproduction

• Steps: call inferTypeFromContext() with a title that both names a subject and starts with an action verb.
• Expected vs Actual:

Title	Actual	Intended
Add dependency upgrade	Feature	Dependency Update
Implement compatibility support	Feature	Compatibility
Add interoperability for runners	Feature	Compatibility
Add integration test for the sync	Feature	Automation
Add GitHub Action for releases	Feature	Automation

• Verified pre-existing: these same titles resolved to Feature on the pre-fix: correct stale issue-type inference and CommonJS title normalisation #3568 source too.

Root Cause

inferMappedValueFromText takes the first rule that both matches and has a configured mapping. type:feature sat second and its patterns include bare action verbs — add, implement, build, develop — that occur in almost any title, so it shadowed every subject rule below it. The specific rules were reachable only from titles that happened to contain no action verb.

Fix Summary

• Move only the generic type:feature rule to the end of the array, leaving every other rule's relative order untouched so the change stays confined to the reported problem.
• Keep type:test above type:compat so integration test remains a test rather than becoming Compatibility.
• Document the ordering invariant in the file so a later rule insertion cannot quietly reintroduce the shadowing.
• This is the ordered-cascade approach — strongest signal first, first match wins — as OpenAlex used when reclassifying 49.6M works: "Rules are ordered by measured trustworthiness: unambiguous structural signals fire first, softer text heuristics later, and the first match wins."

No configuration is touched. project_field_mappings.Type, .github/issue-types.yml and every locked file are unchanged, so derived values remain valid options and nothing downstream has to migrate.

Verification

• Tests added/updated to cover the bug

• Manual verification steps — N/A, no UI; verified by executing the inference directly

• Negative/edge cases checked

• New suite type keyword rules are ordered by signal strength covers all five reported titles plus Add a new endpoint and Implement dark mode (the generic rule must still work), body-level inference, and label precedence.

• Proven to have teeth: reverting only the ordering makes 6 of those 12 cases fail; all pass with the reorder.

• Full suite 282 suites / 5645 tests passed. eslint 0 errors, semgrep 0 findings, markdownlint 0 issues.

• validate:issue-fields passes: cross-file label and issue-type parity intact.

• Changelog gate: 10 failing entries with and without the new entry — 0 new failures.

Risk & Rollback

• Risk level: Low-Medium. This is a deliberate behaviour change to derived values, which is why it is its own PR rather than folded into fix: correct stale issue-type inference and CommonJS title normalisation #3568.
• Titles that change classification are ones naming a specific subject and an action verb; they were previously captured by the generic rule. Add a new endpoint and similar genuinely feature-shaped titles are unaffected.
• An explicit type:* label still beats every keyword rule, so labelled items are unaffected.
• Rollback plan: revert the single commit. No configuration or persisted data is involved.

Changelog

Issue Types Matched by Signal Strength — A title naming a subject now classifies by that subject, so a test or compatibility update is no longer filed as a feature. (#3577)

[coderabbitai]

Warning

Review limit reached

Next included review available in 27 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Repository: lightspeedwp/.github/.coderabbit.yml

Review profile: CHILL

Plan: Advanced

Run ID: 2e0c94cc-3a84-4f3e-8411-4bc8115c86d9

📥 Commits

Reviewing files that changed from the base of the PR and between 22323f8 and c5c6284.

📒 Files selected for processing (3)

• CHANGELOG.md
• scripts/agents/includes/__tests__/derive-project-fields.test.js
• scripts/agents/includes/derive-project-fields.cjs

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

PR Template Routing

Branch Type: fix
Scope: issue-type-rule-specificity-order
Template: pr_bug.md
Labels Applied: type:bug

This PR was automatically routed based on the branch naming strategy.

[qodo-code-review]

PR Summary by Qodo

Order issue-type keyword rules by signal strength

🐞 Bug fix 🧪 Tests 📝 Documentation 🕐 10-20 Minutes

AI Description

• Prioritizes subject-specific issue-type keywords before generic feature action verbs.
• Adds regression coverage for title, body, fallback, and explicit-label precedence.
• Documents the ordering invariant and records the fix in the changelog.

Diagram

graph TD
  C["Issue Context"] --> L{"Explicit Label?"} -- No --> B{"Branch Prefix?"} -- No --> S{"Subject Rule?"} -- No --> F{"Feature Verb?"} -- No --> E["No Inference"]
  L -- Yes --> M["Mapped Type"]
  B -- Yes --> M
  S -- Yes --> M
  F -- Yes --> M

Loading

High-Level Assessment

The ordered-cascade approach is optimal for this targeted fix because it preserves existing first-match behavior, mappings, and precedence while moving only the overly broad feature rule. A weighted or scored classifier could reduce dependence on array order, but would substantially expand scope and alter established inference semantics without clear benefit here.

Files changed (3) +61 / -6

Bug fix (1) +16 / -6

derive-project-fields.cjsMove the generic feature rule behind specific signals +16/-6

Move the generic feature rule behind specific signals

• Moves 'type:feature' to the end of the keyword cascade so subject-specific rules match first. Comments document the ordering invariant and preserve test-before-compatibility behavior.

scripts/agents/includes/derive-project-fields.cjs

Tests (1) +44 / -0

derive-project-fields.test.jsCover signal-strength ordering and precedence +44/-0

Cover signal-strength ordering and precedence

• Adds regression cases for dependency, compatibility, automation, and generic feature titles. It also verifies body inference and confirms explicit labels retain highest precedence.

scripts/agents/includes/tests/derive-project-fields.test.js

Documentation (1) +1 / -0

CHANGELOG.mdDocument corrected issue-type classification +1/-0

Document corrected issue-type classification

• Adds a Fixed entry explaining that subject-bearing titles now avoid incorrect generic Feature classification.

CHANGELOG.md

[github-actions]

📋 Changelog Quality Validation

Metric	Count
✅ Passing	113
❌ Failing	10
🆕 New failures in this PR	0
📦 Pre-existing failures	10

Status

✅ Validation PASSED - No new failures introduced by this PR.
Note: 10 pre-existing failure(s) remain in the Unreleased section.

No action required.

[linear-code]

GIT-2363