Skip to content

Historical field note

This record preserves ground-truth investigation evidence for a single dream finding. It is not a current operating contract; use the numbered book and current runbooks for instructions.

Ground Truth: dream finding dreamrepair:477446f5c8b8bf1972f2ad31444c956b (slack failure_pattern)

Task: Establish the ground truth behind a low-confidence dream-cycle failure_pattern finding, provider label slack, fingerprint dreamrepair:477446f5c8b8bf1972f2ad31444c956b. Retrieve and summarize its single supporting evidence record, identify what the source task actually changed, and map the config surfaces that touch Slack/persona runtime and the openclaw|none constraint. Investigation only — no code, skill, tool, config, or deploy edits.

Sole evidence record: mem_627afb5e4e2c4f09b495eea8984a7b10 (record_type deployment_learning:mac, review_verdict "Convert fleet config to OpenClaw terminology and constrain Slack/persona runtime to openclaw|none"), authored by hub-review-workflow.

Originating source task: task_2a25617199eb43dd8cf95de6c40ef0a1 — "Convert fleet config to OpenClaw terminology and constrain Slack/persona runtime to openclaw|none" (evidence_type=repo_change).

Prepared by: fleet worker (investigation node; no production code, test, skill, config, or deploy edits).

Verdict: NO LIVE DEFECT — the sole evidence is a success record; the "slack" label is incidental

This finding is even weaker than its sibling (dreamrepair:cc1dedb0d3036d289aafc1e42b4a22aa, docs/archive/field-notes/investigation-dreamrepair-cc1dedb0-slack.md). Both trace to the same source task, but this fingerprint's lone supporting record, mem_627afb5e4e2c4f09b495eea8984a7b10, is the terminal hub-review outcome approved_published — the strongest positive signal the pipeline emits — not a failure. There is no failing suite, assertion, trace, or reproducer to repair. The slack provider label is derived verbatim from the source task's title/description wording ("constrain Slack/persona runtime"), not from any Slack transport, chat-gateway runtime, or persona-binding failure.

No files under src/mac/, tests/, skills/, .mac/, or deploy/ were modified by this investigation.

What the sole evidence record actually asserts

mem_627afb5e4e2c4f09b495eea8984a7b10 is the review-stage lesson emitted by the hub when the source task published to main. It is written by _record_review_outcome_lesson (src/mac/services.py:21555) with the mac.deployment_learning.v1 shape:

  • evidence_type: review_verdict
  • outcome: approved_published
  • signals: {"stage": "hub_review"}
  • error_signature: empty string — the emitter sets error_signature="" whenever outcome == "approved_published" (src/mac/services.py:21578). There is no stack trace, no failing test id, and no reproduction; the review_verdict/detail carries only the task title and the publication target.

Because this record encodes a published, approved outcome with an empty error signature, it cannot localize a defect. A failure_pattern keyed on it is a provenance artifact, not a fault.

What the source task actually changed (and it landed intact on main)

task_2a25617199eb43dd8cf95de6c40ef0a1 converted the fleet chat-gateway config to OpenClaw terminology and added a persona/Slack runtime selector allow-list constrained to {openclaw, none}, mirroring the network.provider allow-list pattern. It went running → needs_review → reviewing → approved → published to git://main → completed; the record above is that publication's lesson. The intended change is present in the tree:

  • Persona/Slack runtime selector allow-list — gateway_impl must be openclaw or none, or validation fails loudly (src/mac/fleet_setup.py:208, "gateway_impl must be openclaw or none").
  • OpenClaw defaults read from the openclaw: block with a backward-compatible, read-only hermes: fallback (src/mac/fleet_setup.py:194).
  • Backward-compatible wire format: per-fleet and per-agent output is still emitted under the legacy hermes key (src/mac/fleet_setup.py:295, src/mac/fleet_setup.py:502) so persisted fleets.yaml and the deploy payload contract keep loading.

Config surfaces implicated (Slack/persona runtime + openclaw|none)

  • src/mac/fleet_setup.py — the gateway_impl {openclaw, none} allow-list and openclaw/legacy-hermes defaults resolution; the primary surface the source task edited.
  • src/mac/hermes_config_surface.py — fleet-scoped OpenClaw configuration inspection/apply; retains defaults.hermes schema keys for wire-format compatibility while accepting the OpenClaw-named block first (src/mac/hermes_config_surface.py:9).
  • .mac/project.yaml — project contract that references hub/gateway hosts.
  • src/mac/worker.py — worker-side runtime consumer of the fleet config.
  • src/mac/_hermes/gateway/platforms/ — the vendored chat-gateway transports (including slack), which the finding label points at but which nothing in the evidence implicates.

Why the slack label is incidental, and why confidence is low

The dream classifier's provider matcher includes a bare-word rule (r"\bslack\b", "slack") (src/mac/dream_cycle_classifier.py:143). The source task's title/description contains the word "Slack", so the classifier echoes a slack provider area even though no Slack surface failed. The low confidence (0.35) is the classifier's deterministic single-record structural floor for support < 2 (src/mac/dream_cycle_classifier.py:15, src/mac/dream_cycle_classifier.py:87), i.e. an evidence-volume signal, not a corroborated defect. No skill, tool, or repo_area is co-flagged.

Recurring pattern or isolated event?

Isolated. The finding is backed by exactly one deployment_learning:mac record, and that record is a success (approved_published). Any later re-emissions are the same nap window re-scoring the same lone record, not independent occurrences.

Durable, non-Slack takeaway (out of scope here)

The generic issue is an evidence-provenance gap in the dream pipeline: the failure_pattern classifier consumed a deployment_learning record for a task whose terminal outcome is approved_published and treated the incidental task-title wording ("Slack") as a provider signal. A durable improvement belongs to the classifier/consolidator — do not raise a failure_pattern from a record whose outcome is a terminal success, and prefer the latest terminal outcome per task_id. This is recorded as a follow-up consideration only and is not implemented here, per the directive not to change skills, tools, or config.

Reproduction (read-only, hub queries)

  • mac task show task_2a25617199eb43dd8cf95de6c40ef0a1 --json — shows the approved/published/completed path and the git://main publication.
  • mac admin memory search --task-id task_2a25617199eb43dd8cf95de6c40ef0a1 --json — shows mem_627afb5e4e2c4f09b495eea8984a7b10 (outcome=approved_published, stage=hub_review) among the task's deployment_learning:mac records.