Cairn
NousResearch/hermes-agent

Kimi reasoning token double-counting and compression bugs

Cluster 81481 · agent. Data as of

Nothing open on GitHub. Fix merged in #98998 on 31 Aug 2026. Check it covers your case.

The short answer

0open issues
0open PRs
4items grouped
Known?
Yes. 4 related items grouped here, 1 issue already closed.
Fix event
A fix was merged to main in PR #98998. Check it covers your case. No release data is tracked.
Tracker
The maintainers' tracker marks this unresolved. That is separate from the fix event.
Still open
Nothing open on GitHub.

These items address two related issues with Kimi model handling: reasoning/reasoning_content fields being charged twice in token estimates, and chat histories incorrectly triggering premature compression. The bug report documents the combined ~2.5x estimate inflation and missing replay of reasoning content on the wire.

Open issues0

No open issues.

Open PRs0

No open pull requests.

Fixes on record

Merged to main. A stamp counts the issues a fix is linked to.

  • #98998 fix(compression): compaction dead-loop ended — trigger and tail walks charge stale thinking by wire truth (#84371)

    merged to main

    files: agent/context_compressor.py, agent/conversation_compression.py, agent/conversation_loop.py, agent/message_sanitization.py, agent/model_metadata.py, agent/turn_context.py, tests/agent/test_estimator_parity_84371.py

2 PRs closed without merging

From the close comments: done elsewhere 1 · superseded by another PR 1

  • #81504 fix(compression): stop Kimi histories from triggering premature compression · done elsewhere → 26f178e, 452f6b7, 64cc87e, #98998, #81481, #92690
  • #81509 fix(estimator): charge duplicate reasoning fields once in token estimates · superseded by another PR → #81504
Closed issues · 1
  • Closed#81481[Bug]: reasoning/reasoning_content double-charged on every historical turn, neither replayed on kimi chat/completions wire — residual ~2.5x estimate inflation after #75884
How this cluster was grouped
  • Link evidence (structural links throughout): every member is connected by structural evidence: shared changed files, or a closing/superseded reference.
  • Pairwise link density 0.50: the share of item pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36. 1.00 means every pair is linked.

Pick up a Lab task

Check whether a fix covers the open issues here, with your own agent. Bring back evidence for maintainers.

Open the Lab

At a glance

Subsystem
agent
Merged PRs
#98998
Closed issues
1
Attempts
2 PRs closed without merging
Fix link checked
31 Aug 2026

For your agent

The same facts, typed.

cluster/81481.json
cluster/81481.md