Files
SubMiner/backlog/tasks/task-166 - Harden-SubMiner-change-verification-for-authoritative-agentic-runtime-checks.md
sudacode ee95e86ad5 docs: add stats dashboard design docs, plans, and knowledge base
- Stats dashboard redesign design and implementation plans
- Episode detail and Anki card link design
- Internal knowledge base restructure
- Backlog tasks for testing, verification, and occurrence tracking
2026-03-14 23:11:27 -07:00

4.4 KiB

id, title, status, assignee, created_date, updated_date, labels, dependencies, references, documentation
id title status assignee created_date updated_date labels dependencies references documentation
TASK-166 Harden SubMiner change verification for authoritative agentic runtime checks Done
2026-03-13 05:19 2026-03-13 05:36
testing
agents
verification
/home/sudacode/projects/japanese/SubMiner/.agents/skills/subminer-change-verification/scripts/verify_subminer_change.sh
/home/sudacode/projects/japanese/SubMiner/.agents/skills/subminer-change-verification/scripts/classify_subminer_diff.sh
/home/sudacode/projects/japanese/SubMiner/.agents/skills/subminer-change-verification/SKILL.md
/home/sudacode/projects/japanese/SubMiner/testing-plan.md
/home/sudacode/projects/japanese/SubMiner/docs-site/development.md

Description

Tighten the SubMiner change-verification classifier and verifier so the implementation matches the approved agentic verification plan: authoritative runtime verification must fail closed when unavailable, lane naming should use real-runtime semantics, session and artifact identities must be unique, and the verifier must be safer for parallel agent use.

Acceptance Criteria

  • #1 The verifier uses real-runtime terminology instead of real-gui for authoritative runtime verification
  • #2 Requested authoritative runtime verification fails closed with a non-green outcome when it cannot run, and unknown lanes do not pass open
  • #3 The verifier allocates a unique session identifier and artifact root that does not rely on second-level timestamp uniqueness alone
  • #4 The verifier summary/report output includes explicit top-level status and session metadata needed for agent aggregation
  • #5 The classifier and verifier better reflect runtime-escalation cases for launcher/plugin/socket/runtime-sensitive changes
  • #6 Regression tests cover the new verifier/classifier behavior

Implementation Plan

  1. Add regression tests for classifier/verifier behavior before changing the scripts.
  2. Harden verify_subminer_change.sh to use real-runtime terminology, fail closed for blocked/unknown authoritative verification, and emit unique session metadata in summaries.
  3. Update classify_subminer_diff.sh and the skill doc to use real-runtime escalation language and better flag launcher/plugin/runtime-sensitive paths.
  4. Run targeted regression tests plus a focused dry-run verifier check, then record outcomes and blockers in the task.

Implementation Notes

Added scripts/subminer-change-verification.test.ts to regression-test classifier/verifier behavior around real-runtime naming, fail-closed authoritative verification, unknown lanes, and unique session metadata.

Reworked verify_subminer_change.sh to normalize real-gui to real-runtime, emit unique session IDs and richer summary metadata, block authoritative runtime verification when unavailable, and fail closed for unknown lanes.

Updated classify_subminer_diff.sh to emit real-runtime-candidate for launcher/plugin/runtime-sensitive paths, and updated the active skill doc wording to match the new lane terminology.

Final Summary

Hardened the SubMiner change-verification tooling to match the approved agentic verification plan. The verifier now uses real-runtime terminology for authoritative runtime verification, preserves compatibility with the deprecated real-gui alias, fails closed for unknown lanes, and returns a blocked non-green outcome when requested authoritative runtime verification cannot run. It now allocates a unique session ID and artifact root by default, writes richer session metadata and top-level status into summary.json/summary.txt plus reports/summary.*, and records path selection mode, blockers, and session-local env roots for agent aggregation. The classifier now emits real-runtime-candidate for launcher/plugin/runtime-sensitive paths, and the active skill doc uses the same terminology. Verification ran via bun test scripts/subminer-change-verification.test.ts, direct dry-run smoke checks for blocked real-runtime and failed unknown-lane execution, and a targeted classifier invocation for launcher/plugin paths.