{"repo":"NousResearch/hermes-agent","public_id":"73298","display_label":"Fix reasoning token estimates in preflight compression","summary":"These items address a bug where reasoning_details content was incorrectly included in preflight token estimates, causing compression to trigger too early on thinking models. Fixes also touch compression index clamping, failure cooldown, and user-query preservation.","subsystem":"agent","counts":{"members":4,"open_issues":0,"open_issues_distinct":0,"open_prs":0,"closed_issues":1,"merged_prs":1,"attempt_prs":2},"landed_fix":{"status":"merged_to_main","basis":"structural","pr":75884,"merged_at":"2026-08-01T06:16:59+00:00"},"resolution":{"state":"resolved","by":"pr-75884"},"freshness":{"state":"verified","last_verified_at":"2026-08-01T06:16:59+00:00","is_stale":false},"grouping":{"backbone":"STRUCTURAL_SPANNING","link_density":0.5},"fix_edges":[{"issue_number":73298,"pr_number":75884,"tier":"verified","basis":"closing_ref","confidence":0.95,"recoverable_from_timeline":true,"pr_merged_at":"2026-08-01T06:16:59+00:00"},{"issue_number":74136,"pr_number":75884,"tier":"verified","basis":"closing_ref","confidence":0.95,"recoverable_from_timeline":true,"pr_merged_at":"2026-08-01T06:16:59+00:00"},{"issue_number":75588,"pr_number":75884,"tier":"verified","basis":"closing_ref","confidence":0.95,"recoverable_from_timeline":true,"pr_merged_at":"2026-08-01T06:16:59+00:00"}],"members":[{"kind":"issue","number":73298,"title":"[Bug]: Preflight token estimate counts reasoning_details envelope at chars/4 — compression fires at ~27% of real usage on thinking models","url":null,"state":"closed"},{"kind":"pr","number":73306,"title":"fix: exclude reasoning_details from preflight token estimate (#73298)","url":null,"state":"closed","close_reason":{"refs":[{"kind":"pr","value":"#75884","number":75884},{"kind":"pr","value":"#51800","number":51800},{"kind":"pr","value":"#73298","number":73298}],"category":"merged_via","implication":"superseded"},"merged":false},{"kind":"pr","number":75884,"title":"fix(compression): index clamps, persisted failure cooldown, user-query preservation, thinking-token estimates","url":null,"state":"merged","close_reason":{"category":"merged"},"merged":true,"is_fix_pr":true,"fix_tier":"verified","files_changed":["agent/context_compressor.py","agent/model_metadata.py","contributors/emails/[redacted:email]","gateway/run.py","tests/agent/test_budget_reasoning_details_exclusion.py","tests/agent/test_compressor_tail_cut_oob_fix.py","tests/agent/test_compressor_zero_user_guard.py","tests/gateway/test_session_hygiene.py"]},{"kind":"pr","number":103447,"title":"fix: exclude reasoning_details from preflight token estimate (#73298)","url":null,"state":"closed","close_reason":{"refs":[{"kind":"pr","value":"#73306","number":73306},{"kind":"pr","value":"#75884","number":75884},{"kind":"pr","value":"#73298","number":73298}],"category":"implemented_elsewhere","implication":"superseded"},"merged":false}],"trust":{"default_tier":"verified","includes_suggested":false}}