feat(agent-interface): reconcile per-harness reasoning-effort sets (gtm#398 Part A)#12
Open
vutuanlinh2k2 wants to merge 1 commit into
Open
feat(agent-interface): reconcile per-harness reasoning-effort sets (gtm#398 Part A)#12vutuanlinh2k2 wants to merge 1 commit into
vutuanlinh2k2 wants to merge 1 commit into
Conversation
…th real adapters The chat effort picker offered levels harnesses can't actually run. Add an explicit `harnessReasoningEffortsOverride` keyed to the real cli-bridge adapter sets, preferred by `harnessReasoningEfforts` over the `none…ceiling` slice: - codex → minimal/low/medium/high (drop the inert `none`) - claude-code → low/medium/high/xhigh/ultracode (its `--effort` ladder; ultracode ≈ max) - pi / openclaw → cap at xhigh - kimi-code → binary minimal/high (its `--thinking` on/off toggle) - acp → added to the ignore-effort set (reads no reasoningEffort) The ceiling table now keeps only the no-thinking runners (cli-base/nanoclaw); router/model-driven harnesses (opencode, gemini) still default to the full ladder, narrowed by the model. gtm-agent#398 Part A.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Part A of gtm-agent#398 — make the chat reasoning-effort picker show only what each harness can really run.
What
reasoningEffortsFor/harnessReasoningEffortspreviously sliced the canonical ladder by a per-harnessceiling, which offered levels the cli-bridge adapters don't accept. Add an explicit
harnessReasoningEffortsOverride(preferred over the ceiling slice) keyed to the real adapter sets:minimal·low·medium·highlow·medium·high·xhigh·ultracode(--effortladder; ultracode ≈max)minimal·low·medium·high·xhighminimal·high(binary--thinkingtoggle)Also:
acpadded toharnessIgnoresEffort(its runner reads noreasoningEffort).noneis dropped from the clamp-based harnesses where it's inert (≡auto, which the picker alwaysshows). The ceiling table keeps only the no-thinking runners (
cli-base/nanoclaw); router/model-drivenharnesses (opencode, gemini) keep the full ladder and are narrowed by the model (gtm side, Part C).
Why
Audit found the picker's offered effort levels diverge from what each CLI adapter actually accepts
(codex degrades xhigh→high, claude has no
none/minimal, kimi is binary, etc.). Full write-up ingtm-agent#398.
Verification
tsc --noEmitclean;buildclean; package suite 21/21 (updated per-harness assertions +harnessHonorsEffort("acp") === false).@tangle-network/agent-interfaceminor.Next in the cascade
sandbox-ui consumes this (kimi labels + neutral copy) → gtm bumps both + populates per-model
supportsReasoning. Tracked in gtm-agent#398 (Parts B, C).