Skip to content

EngLoopKit v1.15.3

Choose a tag to compare

View all tags
@stuartpa stuartpa released this 01 Sep 11:56
· 20 commits to master since this release
v1.15.3
93d3ede

Patch release restoring guarded workflows in ordinary active chats.

  • Agents 30 and 31 now activate on each submitted prompt, including after agent selection in an existing chat and after compaction, rather than relying on first-turn-only SessionStart markers.
  • Prompt activation uses an ignored, timestamp/HEAD/stage/session-bound, single-use entry receipt. Rejected, stale, mismatched, corrupt, or replayed activation creates no scope or evidence state.
  • Agent 30 remains read-only except for one create-new schema-valid analysis JSON. Agent 31 still requires a fresh valid analysis plus explicit unique TE-Rxxx approvals and enforces exact prerequisites, paths, commands, source state, and finalized Stop evidence.
  • Stage 21 missing or malformed --incidents / --postmortem context now returns actionable postmortem-context-required recovery instead of terminating chat. No scope or completion is accepted, and PreToolUse denies every tool until exact valid options are resubmitted.
  • Strict postmortem-learning completion validation and all Stage 22 repair gates remain fail closed.
  • Exact surface: 26 commands/agents/prompts; .NET SDK 10.0.303; SEK v0.1.3.
  • Whole-product readiness: PASS. 3,322 generated tests, 264 core/direct tests, and 64 overlay tests passed. All six modules meet at least 95% line and branch coverage; Tool is 97.75% line / 95.00% branch.
  • Final CI, isolated tool install, disposable extension install, agent-surface validation, and clean/coexist overlay transactions: PASS.
Assets 6
Loading