Keeps this page in sync as the body changes. Pause it any time for a quieter view.
Path /specs/attention-heuristics-pipeline-status
Last refresh never
Spec: Attention Heuristics in Pipeline-Status
inventory_source local | spec_api /api/spec-registry/attention-heuristics-pipeline-status | registry_updated 2026-04-09T03:10:08.822699Z
potential_value 0.00 | actual_value 1.00 | value_gap 0.00
estimated_cost 0.00 | actual_cost 1.00 | cost_gap 1.00
estimated_roi 0.00 | actual_roi 1.00
Missing contributor linkage. Submit a change request with contributor attribution.
Open process view for this spec
task_ids -
branches -
source_files specs/agent-orchestration-api.md, specs/attention-heuristics-pipeline-status.md, specs/coherence-network-agent-pipeline.md, specs/pipeline-observability-and-auto-review.md, specs/project-manager-pipeline.md, specs/split-review-deploy-verify-phases.md
evidence_refs -
Open implementation view for this spec
implementation_refs spec-registry:agent-orchestration-api, spec-registry:attention-heuristics-pipeline-status, spec-registry:coherence-network-agent-pipeline, spec-registry:pipeline-observability-and-auto-review, spec-registry:project-manager-pipeline, spec-registry:split-review-deploy-verify-phases
lineage_ids -
public_endpoints -
summary Surface pipeline health signals so operators and automation can detect stuck runs, repeated failures, and low success rate without inspecting logs. The `GET /api/agent/pipeline-status` response includes an `attention` object whose flags drive alerts, Telegram summaries, and optional auto-fix (see spec 027).
process_summary `GET /api/agent/pipeline-status` returns an `attention` object with boolean flags and a `flags` list (spec 002, 027).; Stuck: `attention.stuck` is true when there are pending tasks, no running task, and the longest pending wait time exceeds a thr; Repeated failures: `attention.repeated_failures` is true when the three most recently completed tasks (by completion order) are all failed.; Low success rate: `attention.low_success_rate` is true when metrics over the rolling window (e.g. 7 days) have at least 10 tasks and succe; Heuristic thresholds (stuck minutes, consecutive failure count, success-rate window and minimum sample) are configurable
pseudocode_summary -
implementation_summary api/app/services/agent_service_list.py (get_attention_tasks(), get_review_summary()); api/app/services/agent_service_pipeline_status.py (attention flags, status collection)