[piyushmakhija28]

Closes item 4 of #296, the last of the four.

verify_orchestration_prompt() was written, exported from runtime_verification and covered by four unit tests — and called by nothing. Its sibling verify_orchestrator_result() was deleted in 94cbaf7 for that reason plus a false premise (a str signature for a value that has always been a dict). This one encodes a true contract, so the choice was wire it or delete it. Leaving it was the only option that got worse over time: a correct-but-uncalled validator looks like coverage that does not exist.

Checking first changed the justification

I expected the argument to be "the emitted prompt is unchecked". That was half true and the wrong half. Both degraded paths in the node already log — an ERROR from prompt-gen falls back to the raw task at logger.error, and an empty llm_response/prompt falls back at logger.warning.

The real gap is narrower and, I think, more worth closing: a response that is non-empty, and therefore silent, but not a usable orchestration prompt — a truncated template, the wrong file, a stub. Until now that was indistinguishable from a good prompt.

Not fatal, deliberately

Findings land on orchestrator_result["prompt_warnings"] and are logged. They never fail the step.

STEP1_CONTRACT documents the short raw-task fallback as a legitimate degraded path, and that fallback trips both of the verifier's checks by design. Failing here would convert every recoverable run into a dead one — the same reasoning e0d9372 applied to Phase 2: loud and visible rather than fatal.

Wiring was confirmed possible before it was done

The master template resolved from claude-global-library is 1,996 lines / 335 KB and contains "Phase" 386 times, so the happy path clears both the 200-character floor and the keyword check. Without that check first, this could have logged a spurious warning on every single run.

The test drives the node, not the function

tests/test_schema_verifier.py already covers the function in isolation — and that coverage is exactly what failed to notice it had no callers. So every assertion here goes through step1_task_analysis_node.

The two cases are a pair on purpose. "A good prompt produces no warnings" passes just as well when the verifier is never invoked, because an unwired check and a satisfied check both return an empty list.

Proven by unwiring it: replacing the call with a literal [] turned the malformed-prompt case and the not-fatal case red while the well-formed case stayed green. Then restored. The pair can fail for the reason it claims.

Counts

489 / 101 / 92 / 122 — caught across all seven claim sites by tests/test_doc_counts_match_the_tree.py rather than by hand. Second time that enforcement has paid for itself since it landed two PRs ago.

Verified

78 targeted tests (doc counts, docs index, this wire-in, schema verifier, kg-routing, standards wiring, transition guards), all six CI gates, clean collection, no line-ending flips. ruff is not on PATH in this shell so the lint gate was left to pre-commit and CI, as with #298 where it passed.

🤖 Generated with Claude Code