fix(audit): the harness auditor judged absence from a file it had only partly read (closes #5750)

[vivekchand]

Closes #5750.

#5750 is a false positive, and the cause files more of them

It was filed automatically at severity high: "No code path in the OpenClaw/NemoClaw adapter reads NEMOCLAW_TRACE_FILE / the .e2e/traces directory or parses this trace artifact."

That code has shipped:

clawmetry/adapters/openclaw.py:1705	reads NEMOCLAW_TRACE_FILE, falls back to NEMOCLAW_TRACE_DIR, then the harness default
same file, :1774-1777	sets nemoclawOnboardTraceStatus, nemoclawOnboardTraceSpanCount, nemoclawOnboardTraceErrors, nemoclawOnboardSlowSpans
clawmetry/adapters/nemo.py:1040	reads it too
REQ-OBS-RSO-034	specifies the whole capability, delivered by PR #5198

Why the auditor could not see it

_adapter_source read the first 60,000 characters. openclaw.py is 193,340. The reader sits at line 1690, roughly 18k past the cut — so 69% of the adapter was invisible. Both runtimes the OSS audit covers (openclaw, nemoclaw) map to that same file, so every run judged a two-thirds-clipped adapter.

Then the prompt said:

verify against the FULL adapter above (it is provided in full)

For a task that is entirely about reporting absence, that sentence converts "I did not see it" into "it is not there". The model did what it was told the evidence supported.

This is the second time. The docstring records the first: aider's conditional COST at line ~527 was cut, the audit flagged "no COST", and the cap was raised to 60k in response. Raising a number is not a fix for a file that grows.

The fix

• the budget is far above any adapter here, and _adapter_source returns whether it trimmed rather than trimming silently (the caller also warns);
• when it does trim, the prompt says so instead of claiming completeness;
• _adapter_index always carries every def and every UPPER_CASE string from the whole file, so an absence claim stays checkable even under trimming. Small, complete, and built from the full source regardless of what the body contains.

openclaw  clawmetry/adapters/openclaw.py  chars=193104  trimmed=False
   NEMOCLAW_TRACE_FILE visible in body: True
   index mentions NEMOCLAW_TRACE_FILE:  True

Guard

tests/test_harness_audit_reads_whole_adapter.py auto-discovers from manifest.json, so an adapter that outgrows the budget tomorrow fails here instead of quietly filing fiction. One test asserts the NEMOCLAW_TRACE_FILE witness sits past the old 60k mark, so it keeps proving something.

Proven red — restoring the 60k cap and the old completeness claim:

FAILED test_the_audited_adapter_is_read_whole[openclaw]
FAILED test_the_audited_adapter_is_read_whole[nemoclaw]
FAILED test_the_prompt_never_claims_completeness_it_does_not_have[openclaw]
FAILED test_the_prompt_never_claims_completeness_it_does_not_have[nemoclaw]
FAILED test_the_index_reaches_symbols_past_the_old_60k_cap
5 failed, 1 passed

Blast radius beyond this issue

#5746, #5747, #5748 and #5749 were filed by the same truncated runs. Each needs checking against the full adapter before anyone works it. I have not closed them — that is a per-issue verification, not an assumption — but #5746 already has PR #5760 open against it, which is worth a second look given this.

The cost of this bug is not a wasted CI minute. It is a high-severity issue in the tracker that a person or an agent has to read, reproduce and disprove, against code that already shipped.

No-PRD: bug fix against filed issue #5750.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Xb6A5G74JiMe3zHFs1JZEP

[8090-software-factory]

✅ Drift Bot (ClawMetry): no drift detected

Drift Bot analyzed the changed files against this project's blueprints and requirements and found no drift.

[vivekchand]

Checked the four siblings — they are NOT false positives

I flagged #5746 to #5749 as suspect because they came from the same truncated runs. I checked each against the full adapter, and that caution was too broad. Correcting it here so nobody dismisses four real gaps on the strength of this PR.

The distinction is which region of the file the claim is about. #5750 asserted a specific symbol did not exist, and that symbol sits ~18k past the old cut. The other four reason about functions the model could actually see:

function the audit cited	position	within old 60k cap?
_openclaw_doctor_findings	char 4,899	yes
_clawrouter_detect	char 6,053	yes
_sandbox_inference_configs	char 46,194	yes

Grepping the full clawmetry/adapters/openclaw.py for what each claims is missing:

issue	claim	full-file grep	verdict
#5746	no share/publish state reader	no share_url / is_shared / publish-registry reader; the only share hits are "reasoning-token share", "shared filesystem", "shared plugin-SDK monitor"	genuine gap
#5747	no meeting-library reader	zero hits for meeting	genuine gap
#5748	no connected-accounts / account-priority reader	zero hits	genuine gap
#5749	no update candidate-state / abandoned-update reader	only _gateway_migration_warning (migration warnings from gateway log events), nothing about candidate state or abandoned-update recovery	genuine, narrowed — some migration signal exists, the update-pipeline states do not

So #5750 was the only truncation-caused false positive of the five, and PR #5760 against #5746 is doing real work.

The fix in this PR still stands on its own: the auditor was judging a two-thirds-clipped file while being told it was complete, and that produced one high-severity fiction. It would have produced more as openclaw.py kept growing.

🤖 Generated with Claude Code