fix(agent): adaptive default for agent_input_token_budget (closes #1170) by nickorlabs · Pull Request #1190 · odysseus-dev/odysseus

nickorlabs · 2026-06-02T14:19:07Z

Summary

Closes #1170.

Replaces the flat agent_input_token_budget = 6000 default with adaptive semantics so users on long-context models (kimi-k2:1t, minimax-m3, gpt-oss 128K, etc.) actually use the context window they have. Sibling to the now-merged #1056 — that PR made Ollama honor num_ctx; this one makes the app-side budget meaningful in the first place.

Semantics

`agent_input_token_budget` value	Effective behavior
unset / `None` / `""` / `"auto"` (the new default)	adaptive: `min(context_length * 0.85, agent_input_token_hard_max)`
`0`	explicit "no soft trim" — preserves the existing escape hatch
`>0`	explicit cap, bounded by `context_length` when known
negative / malformed string	treated as unset → auto

agent_input_token_hard_max (new, default 200_000) caps the auto branch only — explicit positive budgets are not capped by it (user knows what they want).

What this PR changes

src/agent_loop.py
- DEFAULT_INPUT_TOKEN_HARD_MAX = 200_000 (module-level, importable)
- _default_input_budget(ctx, hard_max) — pure helper
- _resolve_input_token_budget(ctx) — full semantic resolver
- call site swapped to use the resolver
src/settings.py — both keys registered in DEFAULT_SETTINGS so the normal settings paths (/api/auth/settings, the manage_settings tool) can read/persist them:
- agent_input_token_budget default 6000 → "auto"
- agent_input_token_hard_max added with default 200_000
- Each entry has a multi-line comment inline.
src/tool_implementations.py — manage_settings friendly aliases extended:
- existing "token budget" → agent_input_token_budget (kept)
- new "input budget" → agent_input_token_budget
- new "hard max", "token budget cap", "input budget cap" → agent_input_token_hard_max

Backwards compatibility

Existing state	Now
`data/settings.json` has no `agent_input_token_budget` key	Fresh `"auto"` from `DEFAULT_SETTINGS` → adaptive default kicks in
`data/settings.json` has explicit `6000` (or any positive int)	Unchanged — resolver treats as explicit cap, returns the same number
`data/settings.json` has explicit `0`	Unchanged — no soft trim
`data/settings.json` has explicit `"auto"`	Adaptive default

Pinned by regression tests for the 6000-preserved and 0-preserved cases.

Tests

83 total, all passing (49 pre-existing + 8 pure-helper + 17 resolver + 6 DEFAULT_SETTINGS-registration + 3 alias-registration).

The TestDefaultSettingsRegistration group exercises the end-to-end auto-default-when-unset path against the real load_settings (using tmp_path so it's hermetic), not just the mocked resolver case. The alias tests grep src/tool_implementations.py directly to avoid pulling the full app stack but still pin that the friendly names exist.

$ python -m pytest tests/test_agent_loop.py -v
======================== 83 passed in 0.18s ========================

Diff

 src/agent_loop.py             |  76 +++++++++++++++++++++++++++++-
 src/settings.py               |  19 +++++++-
 src/tool_implementations.py   |   4 +-
 tests/test_agent_loop.py      | 245 ++++++++++++++++++++++++++++++++++++++-
 4 files changed, 341 insertions(+), 3 deletions(-)

Adjacent context

Proposal: make agent_input_token_budget adapt to model context window (default 6000 silently caps long-context models) #1170 — the spec issue this closes
fix(llm): thread num_ctx into Ollama requests (#909) #1056 (merged) — num_ctx now correctly sent to Ollama; combined, the full long-context pipeline works
fix(model_context): query Ollama /api/show for actual serving context #753 (open) — how Odysseus discovers context_length; orthogonal to this PR

pewdiepie-archdaemon · 2026-06-02T14:22:09Z

This is closer because it adds a hard ceiling, but I would not merge this shape yet.

Two blockers:

agent_input_token_budget = 0 currently means no soft trim. On main, soft_budget = int(get_setting(..., 6000) or 0) followed by if soft_budget > 0: skips trimming when the setting is explicitly 0. This PR changes 0 to auto, so any deployment that intentionally disabled soft trimming would start trimming again.
The hard max is a function-local constant. For this setting, the ceiling should be configurable or at least represented as a named setting/default with tests. The original issue explicitly called out surprise cost/latency risk on premium APIs, and a hidden constant is hard for users/admins to reason about.

Suggested semantics:

unset/default: auto with bounded hard max
0: preserve existing unlimited/no-soft-trim behavior
>0: explicit cap, still bounded by context length
auto hard max: configurable default, covered by tests

Also please add/adjust tests for the explicit 0 case so we do not regress existing configs.

@pewdiepie-archdaemon

…sseus-dev#1170) The flat agent_input_token_budget=6000 default soft-capped every agent chat at 6000 input tokens regardless of model capability. Combined with the (now-fixed) Ollama num_ctx silent 2K cap, this meant users on long-context models (kimi-k2:1t, minimax-m3 at 1M, gpt-oss at 128K) silently lost most of their advertised context window — and the pasted-message head+tail truncation in src/context_compactor.py was the visible symptom. Semantics (revised per odysseus-dev#1190 review feedback from @pewdiepie-archdaemon): unset / None / empty / "auto" -> adaptive default, capped at the agent_input_token_hard_max setting (default ceiling: 200_000) 0 -> explicit "no soft trim" — preserves the existing escape-hatch for deployments that turned trimming off >0 -> explicit cap; still bounded by context_length when known so the budget never exceeds the real window negative / malformed string -> treated as unset (auto) Configurable hard max: The 200_000 ceiling is now a module-level DEFAULT_INPUT_TOKEN_HARD_MAX constant AND a tunable agent_input_token_hard_max setting. Admins can raise it (premium-API users with very large context) or lower it (cost-paranoid deployments) without code changes. Default chosen as "roomy but not runaway" for premium APIs. Three new symbols: - DEFAULT_INPUT_TOKEN_HARD_MAX module-level constant - _default_input_budget(ctx, hm) pure helper: ctx -> min(0.85*ctx, hm) - _resolve_input_token_budget(ctx) full semantic resolver Tests: - 74 total, all passing (49 pre-existing + 8 helper + 17 resolver) - explicit 0 case has 3 tests (int 0, string "0", and 0 with unknown ctx) to lock in the no-soft-trim escape hatch - the agent_input_token_hard_max setting has 4 tests (raise, lower, 0 falls back to default, malformed falls back to default) - a regression test pinning that an explicitly-stored agent_input_token_budget=6000 still yields exactly 6000 (existing user configs are preserved verbatim) Empirical evidence from a real Linux/Docker test deployment that motivated this: src.model_context - INFO - Context length for minimax-m3: 1000000 src.context_compactor - INFO - Trimming messages: 51967 > 4976 budget (ctx=6000) src.agent_loop - INFO - [agent] soft-trimmed context: 51967 -> 7123 tokens (budget=6000, reserve=1024) ctx=1M was discovered correctly, then min(1M, 6000) = 6000 kicked in. With this PR's adaptive default, the same setup yields a 200_000 budget, which the user can raise via agent_input_token_hard_max if they want the full window.

nickorlabs · 2026-06-02T14:29:47Z

Thanks for the careful read — both blockers were real. Pushed a revision (force-with-lease) that implements your suggested semantics.

What changed

Setting value	Old PR behavior	New behavior
unset / `None` / `""` / `"auto"`	adaptive (function-local 200K cap)	adaptive, cap from `agent_input_token_hard_max` setting
`0`	adaptive (regression — broke existing configs)	explicit disable, no soft trim (preserves existing escape hatch)
`>0`	unbounded explicit cap	explicit cap, bounded by `context_length` when known
negative / malformed string	error / silent 6000	treated as unset → auto

Configurable hard max

HARD_MAX is now DEFAULT_INPUT_TOKEN_HARD_MAX = 200_000 at module scope and overridable via a new agent_input_token_hard_max setting. Lower it for cost paranoia, raise it for premium-API setups with very large context. Tests cover raise, lower, zero-falls-back-to-default, and malformed-falls-back-to-default.

Three symbols now

DEFAULT_INPUT_TOKEN_HARD_MAX — module-level constant, importable, replaces the function-local literal
_default_input_budget(ctx, hard_max) — pure helper, no settings IO
_resolve_input_token_budget(ctx) — full semantic resolver that reads both settings

The split lets the resolver be tested with monkeypatch.setattr(agent_loop, "get_setting", ...) (no full settings stack stand-up) and lets the pure helper be tested with literal values.

Tests

49 pre-existing + 8 helper + 17 resolver = 74 total, all pass.

Explicit-0 case has three pinned tests (int 0, string "0", and 0 with unknown ctx) so the "no soft trim" escape hatch can't silently regress. There's also a regression test confirming that an existing user config of literal 6000 still yields exactly 6000.

$ python -m pytest tests/test_agent_loop.py -v
======================== 74 passed in 0.17s ========================

Ready for another look when you have a minute.

pewdiepie-archdaemon · 2026-06-02T14:32:34Z

The revised resolver is much better and preserves the 0 no-soft-trim escape hatch. One remaining blocker: agent_input_token_hard_max is not actually configurable through the normal settings path yet.

/api/auth/settings only persists keys that exist in DEFAULT_SETTINGS, and this PR only reads get_setting("agent_input_token_hard_max", DEFAULT_INPUT_TOKEN_HARD_MAX) without adding agent_input_token_hard_max to src/settings.py. That means admins cannot save the hard max through the standard settings API, so the “configurable hard max” claim is only true if someone manually edits data/settings.json or another code path writes an unknown key.

Please add the setting to DEFAULT_SETTINGS with a comment, and ideally expose/document it alongside the input budget setting or through the existing settings-management aliases. Also update the PR body, since it still describes the old version where 0 meant auto and the hard max was function-local.

…sseus-dev#1170) Replaces the flat agent_input_token_budget=6000 default with adaptive semantics so users on long-context models (kimi-k2:1t, minimax-m3, gpt-oss at 128K, etc.) actually use the context window they have. Combined with the (now-merged) odysseus-dev#1056 Ollama num_ctx fix, this is the second half of the end-to-end long-context UX: odysseus-dev#1056 made Ollama honor the budget, this PR makes the budget meaningful. ## Semantics (per maintainer review on odysseus-dev#1190) unset / None / empty / "auto" -> adaptive default: ~85% of context_length, capped at the agent_input_token_hard_max setting 0 -> explicit "no soft trim" — preserves the existing escape hatch for deployments that turned trimming off >0 -> explicit cap, bounded by context_length when known negative / malformed string -> treated as unset (auto) ## What's in this PR 1. src/agent_loop.py - DEFAULT_INPUT_TOKEN_HARD_MAX = 200_000 (module-level, importable) - _default_input_budget(ctx, hard_max) — pure helper - _resolve_input_token_budget(ctx) — full semantic resolver - call site swapped to use the resolver 2. src/settings.py — both keys registered in DEFAULT_SETTINGS so the normal settings-management paths (/api/auth/settings, manage_settings tool) can persist them: - agent_input_token_budget default changed from 6000 to "auto" - agent_input_token_hard_max added with default 200_000 Each entry has a multi-line comment explaining the semantics inline. 3. src/tool_implementations.py — manage_settings friendly aliases: - existing 'token budget' -> agent_input_token_budget (kept) - new 'input budget' -> agent_input_token_budget - new 'hard max' -> agent_input_token_hard_max - new 'token budget cap' / 'input budget cap' -> agent_input_token_hard_max ## Backwards compatibility Users with an explicitly-persisted agent_input_token_budget=6000 in their data/settings.json keep exactly 6000 (the resolver treats it as an explicit cap). Pinned by a regression test. Fresh installs and users who never touched the setting get the adaptive default — that's the UX win. Users who explicitly stored 0 keep 0 (no soft trim), also pinned by tests. ## Tests 83 total, all passing (49 pre-existing + 8 helper + 17 resolver + 6 DEFAULT_SETTINGS-registration + 3 alias-registration). The DEFAULT_SETTINGS tests use real load_settings against a temp file so the end-to-end auto-default-when-unset path is exercised, not just the mocked-resolver case. python -m pytest tests/test_agent_loop.py -v ======================== 83 passed in 0.18s ======================== ## Empirical evidence (motivation) src.model_context - INFO - Context length for minimax-m3: 1000000 src.context_compactor - INFO - Trimming messages: 51967 > 4976 budget (ctx=6000) src.agent_loop - INFO - [agent] soft-trimmed context: 51967 -> 7123 tokens (budget=6000, reserve=1024) ctx=1M was discovered correctly, then min(1M, 6000) = 6000 dominated. With this PR + default 'auto' + hard_max=200_000, the same setup yields a 200K budget out of the box; admins who want more raise the hard_max.

nickorlabs · 2026-06-02T14:59:17Z

Good catch — agent_input_token_hard_max was reachable only via direct file edit before. Fixed in ee57701:

What changed in this revision

DEFAULT_SETTINGS registration in src/settings.py — both keys now persistable through /api/auth/settings and the manage_settings tool:
- agent_input_token_budget default changed from 6000 to "auto" (the sentinel that routes through the resolver's auto branch)
- agent_input_token_hard_max added with default 200_000
- Each entry has a multi-line comment explaining the semantics inline so future readers don't have to dig
Friendly aliases in src/tool_implementations.py for the manage_settings tool:
- "hard max" → agent_input_token_hard_max
- "token budget cap" / "input budget cap" → agent_input_token_hard_max
- new "input budget" alias for the existing agent_input_token_budget (pair with the new alias above)
- existing "token budget" kept
PR body rewritten to describe the current shape — old "old → new" comparison table is gone, replaced with the semantic table and the registration story.
6 new DEFAULT_SETTINGS-registration tests + 3 alias tests:
- agent_input_token_budget is in DEFAULT_SETTINGS and defaults to "auto"
- agent_input_token_hard_max is in DEFAULT_SETTINGS with the expected value
- Brand-new install (no data/settings.json) routes through the auto branch via the real load_settings against a temp file (hermetic)
- User who explicitly saved 6000 keeps exactly 6000 (regression pin, real load_settings)
- All three new friendly aliases are registered

83 passed in 0.18s overall (tests/test_agent_loop.py). Branch HEAD is ee57701, MERGEABLE / CLEAN.

Anything else you'd like me to adjust?

pewdiepie-archdaemon · 2026-06-02T15:02:26Z

The updated semantics and settings registration address my earlier blockers: 0 remains no-soft-trim, the hard max is in DEFAULT_SETTINGS, and the PR body now matches the current behavior.

But the branch is stale relative to current main. After git fetch origin main, origin/main..HEAD includes a large set of unrelated already-merged files (app.py, auth/routes/model/search/readiness/url-safety/session-sidebar/Cookbook files, many tests, etc.), not just the agent budget changes.

Please rebase/update so the two-dot diff contains only the intended files:

src/agent_loop.py
src/settings.py
src/tool_implementations.py
tests/test_agent_loop.py

Once clean, this is ready for focused test review with ./venv/bin/python -m pytest -q tests/test_agent_loop.py and git diff --check origin/main...HEAD.

pewdiepie-archdaemon · 2026-06-02T15:14:11Z

Closing as superseded by #1230, which has now merged and closed #1170. Thanks for iterating on the semantics here; the final merged fix keeps the focused scope clean and preserves the existing 0 no-soft-trim behavior while adapting the default to the discovered context window.

…_token_hard_max setting Completes the reviewer requirement from PR odysseus-dev#1190 review that was carried over but not implemented in odysseus-dev#1230: > "The hard max is a function-local constant. For this setting, the ceiling > should be configurable or at least represented as a named setting/default > with tests." — review on odysseus-dev#1190 odysseus-dev#1230 shipped the adaptive auto-derivation but left `DEFAULT_HARD_MAX = 200_000` as a hardcoded module constant in src/context_budget.py. Admins on premium APIs with large context windows (kimi-k2 / minimax-m3 at 1M, etc.) can use their full window today only by setting `agent_input_token_budget` explicitly — which then takes them off the adaptive auto-path entirely. ## What this PR changes - src/settings.py: register `agent_input_token_hard_max` in DEFAULT_SETTINGS, default 200_000 (matches `DEFAULT_HARD_MAX`). Inline comment documents the no-op semantics in the explicit branch. - src/agent_loop.py: read the setting at the call site and pass it as the `hard_max` kwarg of `compute_input_token_budget`. Defensive parsing — missing / non-int / zero values fall back to `DEFAULT_HARD_MAX`, so a misconfig cannot silently zero the budget. - src/tool_implementations.py: three friendly aliases for `manage_settings`: - "hard max" -> agent_input_token_hard_max - "token budget cap" -> agent_input_token_hard_max - "input budget cap" -> agent_input_token_hard_max Plus the existing "token budget" -> agent_input_token_budget keeps a matching shorter alias "input budget". - tests/test_context_budget.py: 6 new tests on top of the existing 6: - hard_max raises the auto ceiling (1M ctx + raised cap -> 85% of ctx) - hard_max lowers the auto ceiling (128K ctx + 50K cap -> 50K) - hard_max has no effect on the explicit branch - DEFAULT_SETTINGS contains the new key - manage_settings aliases are registered - the live get_setting path returns the override value, and malformed values fall back per the agent_loop defensive parsing 12 passed in 0.04s. No changes to the pure helper signature or semantics; odysseus-dev#1230's behavior is the default when the new setting is unset. ## How it lets users drop the explicit override Before this PR, on a 1M-context model: agent_input_token_budget = 900_000 (explicit) -> 900K [user override] agent_input_token_budget = <unset> (auto) -> 200K [HARD_MAX] After this PR, same model: agent_input_token_budget = <unset> agent_input_token_hard_max = 900_000 -> min(1M * 0.85, 900K) = 850K [auto, no override needed] The explicit-override path keeps working unchanged for users who prefer it.

…_token_hard_max setting (#1273) Completes the reviewer requirement from PR #1190 review that was carried over but not implemented in #1230: > "The hard max is a function-local constant. For this setting, the ceiling > should be configurable or at least represented as a named setting/default > with tests." — review on #1190 #1230 shipped the adaptive auto-derivation but left `DEFAULT_HARD_MAX = 200_000` as a hardcoded module constant in src/context_budget.py. Admins on premium APIs with large context windows (kimi-k2 / minimax-m3 at 1M, etc.) can use their full window today only by setting `agent_input_token_budget` explicitly — which then takes them off the adaptive auto-path entirely. ## What this PR changes - src/settings.py: register `agent_input_token_hard_max` in DEFAULT_SETTINGS, default 200_000 (matches `DEFAULT_HARD_MAX`). Inline comment documents the no-op semantics in the explicit branch. - src/agent_loop.py: read the setting at the call site and pass it as the `hard_max` kwarg of `compute_input_token_budget`. Defensive parsing — missing / non-int / zero values fall back to `DEFAULT_HARD_MAX`, so a misconfig cannot silently zero the budget. - src/tool_implementations.py: three friendly aliases for `manage_settings`: - "hard max" -> agent_input_token_hard_max - "token budget cap" -> agent_input_token_hard_max - "input budget cap" -> agent_input_token_hard_max Plus the existing "token budget" -> agent_input_token_budget keeps a matching shorter alias "input budget". - tests/test_context_budget.py: 6 new tests on top of the existing 6: - hard_max raises the auto ceiling (1M ctx + raised cap -> 85% of ctx) - hard_max lowers the auto ceiling (128K ctx + 50K cap -> 50K) - hard_max has no effect on the explicit branch - DEFAULT_SETTINGS contains the new key - manage_settings aliases are registered - the live get_setting path returns the override value, and malformed values fall back per the agent_loop defensive parsing 12 passed in 0.04s. No changes to the pure helper signature or semantics; #1230's behavior is the default when the new setting is unset. ## How it lets users drop the explicit override Before this PR, on a 1M-context model: agent_input_token_budget = 900_000 (explicit) -> 900K [user override] agent_input_token_budget = <unset> (auto) -> 200K [HARD_MAX] After this PR, same model: agent_input_token_budget = <unset> agent_input_token_hard_max = 900_000 -> min(1M * 0.85, 900K) = 850K [auto, no override needed] The explicit-override path keeps working unchanged for users who prefer it.

…_token_hard_max setting (odysseus-dev#1273) Completes the reviewer requirement from PR odysseus-dev#1190 review that was carried over but not implemented in odysseus-dev#1230: > "The hard max is a function-local constant. For this setting, the ceiling > should be configurable or at least represented as a named setting/default > with tests." — review on odysseus-dev#1190 odysseus-dev#1230 shipped the adaptive auto-derivation but left `DEFAULT_HARD_MAX = 200_000` as a hardcoded module constant in src/context_budget.py. Admins on premium APIs with large context windows (kimi-k2 / minimax-m3 at 1M, etc.) can use their full window today only by setting `agent_input_token_budget` explicitly — which then takes them off the adaptive auto-path entirely. ## What this PR changes - src/settings.py: register `agent_input_token_hard_max` in DEFAULT_SETTINGS, default 200_000 (matches `DEFAULT_HARD_MAX`). Inline comment documents the no-op semantics in the explicit branch. - src/agent_loop.py: read the setting at the call site and pass it as the `hard_max` kwarg of `compute_input_token_budget`. Defensive parsing — missing / non-int / zero values fall back to `DEFAULT_HARD_MAX`, so a misconfig cannot silently zero the budget. - src/tool_implementations.py: three friendly aliases for `manage_settings`: - "hard max" -> agent_input_token_hard_max - "token budget cap" -> agent_input_token_hard_max - "input budget cap" -> agent_input_token_hard_max Plus the existing "token budget" -> agent_input_token_budget keeps a matching shorter alias "input budget". - tests/test_context_budget.py: 6 new tests on top of the existing 6: - hard_max raises the auto ceiling (1M ctx + raised cap -> 85% of ctx) - hard_max lowers the auto ceiling (128K ctx + 50K cap -> 50K) - hard_max has no effect on the explicit branch - DEFAULT_SETTINGS contains the new key - manage_settings aliases are registered - the live get_setting path returns the override value, and malformed values fall back per the agent_loop defensive parsing 12 passed in 0.04s. No changes to the pure helper signature or semantics; odysseus-dev#1230's behavior is the default when the new setting is unset. ## How it lets users drop the explicit override Before this PR, on a 1M-context model: agent_input_token_budget = 900_000 (explicit) -> 900K [user override] agent_input_token_budget = <unset> (auto) -> 200K [HARD_MAX] After this PR, same model: agent_input_token_budget = <unset> agent_input_token_hard_max = 900_000 -> min(1M * 0.85, 900K) = 850K [auto, no override needed] The explicit-override path keeps working unchanged for users who prefer it.

Address RaresKeY's review: P2 (explicitness): is_setting_customized treated a saved value equal to the default as "not explicit", which ALSO blocked a user from deliberately pinning the default budget. Reframe the default value itself as the AUTO sentinel — agent_input_token_budget == DEFAULT_BUDGET means "scale to the model's context window", any other value is an explicit cap. A materialized default still reads as auto (fixing the original regression), and any non-default value the user chooses is now honoured. Drop the now-unused is_setting_customized helper. P2 (fallback context): auto-scaling trusted get_context_length() even when it returned only the bare DEFAULT_CONTEXT fallback (no endpoint-reported / known window), over-allocating on self-hosted/proxy setups. Add get_context_length_known() (also returns whether the window was actually discovered); the budget block passes 0 when unknown so auto-scaling stays conservative instead of inflating to an unproven window. hard_max stays auto-only — a deliberate explicit budget wins (odysseus-dev#1190); kept that contract and answered the reviewer's question rather than silently reversing it. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…iew odysseus-dev#4122) - settings.py: an explicit budget is clamped to the window only — hard_max is auto-only (odysseus-dev#1190); drop the incorrect "and to hard_max". - is_setting_overridden docstring: drop the stale "adaptive budgets" example; point value-sensitive callers at context_budget.budget_is_explicit. - Tighten the budget-block comments to the contract (default = auto sentinel, non-default = explicit cap, hard_max = auto-only ceiling). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

odysseus-dev#1230/odysseus-dev#1273) The context-budget contract (auto-sentinel, explicit budgets honoured, hard_max auto-only) merged via odysseus-dev#1230 — odysseus-dev#1190 was the earlier, closed, superseded PR. Re-point the contract comments at odysseus-dev#1230 (the live source, already cited for the auto-sentinel two lines up in settings.py). The configurable hard_max setting (`agent_input_token_hard_max`) was a reviewer requirement first raised on odysseus-dev#1190, omitted from the merged odysseus-dev#1230, and actually added in odysseus-dev#1273 — credit odysseus-dev#1273 for it and correct the test comment's history (it previously implied this PR completed the requirement). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ndow scaling (#4122) * fix(agent): don't let a materialized default budget defeat context scaling #1230 scales agent_input_token_budget to the model's context window unless the user explicitly set a budget, detected via is_setting_overridden(). But the settings-save path materializes every DEFAULT_SETTINGS key into settings.json (load_settings merges defaults; handlers persist the merged dict), so the persisted default 6000 reads as "overridden" and the budget code takes the min(6000, ctx) branch — silently re-capping long-context models at 6000 for anyone who has ever saved a setting. This reintroduces the exact regression #1170/#1230 set out to fix. Add is_setting_customized() (saved value != default) and gate the scaling on it instead of mere presence. A persisted default is not a user choice. is_setting_overridden has exactly one consumer (this budget path), so the change is contained. Tests cover the materialized-default regression, a deliberately-chosen budget still being honoured, and the absent-key case. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(agent): rework context-budget fix per review (#4122) Address RaresKeY's review: P2 (explicitness): is_setting_customized treated a saved value equal to the default as "not explicit", which ALSO blocked a user from deliberately pinning the default budget. Reframe the default value itself as the AUTO sentinel — agent_input_token_budget == DEFAULT_BUDGET means "scale to the model's context window", any other value is an explicit cap. A materialized default still reads as auto (fixing the original regression), and any non-default value the user chooses is now honoured. Drop the now-unused is_setting_customized helper. P2 (fallback context): auto-scaling trusted get_context_length() even when it returned only the bare DEFAULT_CONTEXT fallback (no endpoint-reported / known window), over-allocating on self-hosted/proxy setups. Add get_context_length_known() (also returns whether the window was actually discovered); the budget block passes 0 when unknown so auto-scaling stays conservative instead of inflating to an unproven window. hard_max stays auto-only — a deliberate explicit budget wins (#1190); kept that contract and answered the reviewer's question rather than silently reversing it. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * test(agent): lock the materialized-default budget regression (review on #4121) Per WGlynn's review on the issue: add an end-to-end regression that saves an UNRELATED setting (which makes the settings-save path materialize the budget default into settings.json) and asserts the budget still auto-scales rather than re-reading as an explicit 6000 cap — locking the exact reopening shut. To make the test bite the production decision (not just re-derive it), extract `budget_is_explicit()` into src/context_budget.py and use it from the agent loop. It keys off value-vs-default (the default is the auto sentinel), NOT settings presence — which is the whole point, since the save path materializes defaults. Note: after this PR's rework, is_setting_overridden has ZERO production callers, so the merged-dict materialization smell can't reach any setting through a presence check today (WGlynn's durability concern). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(agent): bind the budget context window to its own provenance (review #4122) RaresKeY caught a correctness bug in the fallback-context guard: stream_agent_loop kept only the `known` flag from get_context_length_known() and budgeted off the passed-in `context_length`, which can come from a *different* lookup. Two failures: - local endpoints are re-queried, so the passed value can be a stale DEFAULT_CONTEXT fallback while the fresh probe proves the real (smaller) served context — we'd scale off the stale value; - callers that don't pass context_length (scheduled tasks, teacher escalation, skill test runs, bg_monitor) were capped at 6000 even when a long window is discoverable. Extract budget_context_for_model() which returns the freshly-probed window when known else 0, binding the flag to the value it proves; the agent loop uses it. Regression tests cover the stale-fallback, no-arg-caller, and probe-error paths. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(agent): fix stale budget comments + tighten to the contract (review #4122) - settings.py: an explicit budget is clamped to the window only — hard_max is auto-only (#1190); drop the incorrect "and to hard_max". - is_setting_overridden docstring: drop the stale "adaptive budgets" example; point value-sensitive callers at context_budget.budget_is_explicit. - Tighten the budget-block comments to the contract (default = auto sentinel, non-default = explicit cap, hard_max = auto-only ceiling). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(agent): correct budget issue citations (#1190 → merged #1230/#1273) The context-budget contract (auto-sentinel, explicit budgets honoured, hard_max auto-only) merged via #1230 — #1190 was the earlier, closed, superseded PR. Re-point the contract comments at #1230 (the live source, already cited for the auto-sentinel two lines up in settings.py). The configurable hard_max setting (`agent_input_token_hard_max`) was a reviewer requirement first raised on #1190, omitted from the merged #1230, and actually added in #1273 — credit #1273 for it and correct the test comment's history (it previously implied this PR completed the requirement). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ndow scaling (odysseus-dev#4122) * fix(agent): don't let a materialized default budget defeat context scaling odysseus-dev#1230 scales agent_input_token_budget to the model's context window unless the user explicitly set a budget, detected via is_setting_overridden(). But the settings-save path materializes every DEFAULT_SETTINGS key into settings.json (load_settings merges defaults; handlers persist the merged dict), so the persisted default 6000 reads as "overridden" and the budget code takes the min(6000, ctx) branch — silently re-capping long-context models at 6000 for anyone who has ever saved a setting. This reintroduces the exact regression odysseus-dev#1170/odysseus-dev#1230 set out to fix. Add is_setting_customized() (saved value != default) and gate the scaling on it instead of mere presence. A persisted default is not a user choice. is_setting_overridden has exactly one consumer (this budget path), so the change is contained. Tests cover the materialized-default regression, a deliberately-chosen budget still being honoured, and the absent-key case. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(agent): rework context-budget fix per review (odysseus-dev#4122) Address RaresKeY's review: P2 (explicitness): is_setting_customized treated a saved value equal to the default as "not explicit", which ALSO blocked a user from deliberately pinning the default budget. Reframe the default value itself as the AUTO sentinel — agent_input_token_budget == DEFAULT_BUDGET means "scale to the model's context window", any other value is an explicit cap. A materialized default still reads as auto (fixing the original regression), and any non-default value the user chooses is now honoured. Drop the now-unused is_setting_customized helper. P2 (fallback context): auto-scaling trusted get_context_length() even when it returned only the bare DEFAULT_CONTEXT fallback (no endpoint-reported / known window), over-allocating on self-hosted/proxy setups. Add get_context_length_known() (also returns whether the window was actually discovered); the budget block passes 0 when unknown so auto-scaling stays conservative instead of inflating to an unproven window. hard_max stays auto-only — a deliberate explicit budget wins (odysseus-dev#1190); kept that contract and answered the reviewer's question rather than silently reversing it. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * test(agent): lock the materialized-default budget regression (review on odysseus-dev#4121) Per WGlynn's review on the issue: add an end-to-end regression that saves an UNRELATED setting (which makes the settings-save path materialize the budget default into settings.json) and asserts the budget still auto-scales rather than re-reading as an explicit 6000 cap — locking the exact reopening shut. To make the test bite the production decision (not just re-derive it), extract `budget_is_explicit()` into src/context_budget.py and use it from the agent loop. It keys off value-vs-default (the default is the auto sentinel), NOT settings presence — which is the whole point, since the save path materializes defaults. Note: after this PR's rework, is_setting_overridden has ZERO production callers, so the merged-dict materialization smell can't reach any setting through a presence check today (WGlynn's durability concern). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(agent): bind the budget context window to its own provenance (review odysseus-dev#4122) RaresKeY caught a correctness bug in the fallback-context guard: stream_agent_loop kept only the `known` flag from get_context_length_known() and budgeted off the passed-in `context_length`, which can come from a *different* lookup. Two failures: - local endpoints are re-queried, so the passed value can be a stale DEFAULT_CONTEXT fallback while the fresh probe proves the real (smaller) served context — we'd scale off the stale value; - callers that don't pass context_length (scheduled tasks, teacher escalation, skill test runs, bg_monitor) were capped at 6000 even when a long window is discoverable. Extract budget_context_for_model() which returns the freshly-probed window when known else 0, binding the flag to the value it proves; the agent loop uses it. Regression tests cover the stale-fallback, no-arg-caller, and probe-error paths. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(agent): fix stale budget comments + tighten to the contract (review odysseus-dev#4122) - settings.py: an explicit budget is clamped to the window only — hard_max is auto-only (odysseus-dev#1190); drop the incorrect "and to hard_max". - is_setting_overridden docstring: drop the stale "adaptive budgets" example; point value-sensitive callers at context_budget.budget_is_explicit. - Tighten the budget-block comments to the contract (default = auto sentinel, non-default = explicit cap, hard_max = auto-only ceiling). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(agent): correct budget issue citations (odysseus-dev#1190 → merged odysseus-dev#1230/odysseus-dev#1273) The context-budget contract (auto-sentinel, explicit budgets honoured, hard_max auto-only) merged via odysseus-dev#1230 — odysseus-dev#1190 was the earlier, closed, superseded PR. Re-point the contract comments at odysseus-dev#1230 (the live source, already cited for the auto-sentinel two lines up in settings.py). The configurable hard_max setting (`agent_input_token_hard_max`) was a reviewer requirement first raised on odysseus-dev#1190, omitted from the merged odysseus-dev#1230, and actually added in odysseus-dev#1273 — credit odysseus-dev#1273 for it and correct the test comment's history (it previously implied this PR completed the requirement). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…_token_hard_max setting (odysseus-dev#1273) Completes the reviewer requirement from PR odysseus-dev#1190 review that was carried over but not implemented in odysseus-dev#1230: > "The hard max is a function-local constant. For this setting, the ceiling > should be configurable or at least represented as a named setting/default > with tests." — review on odysseus-dev#1190 odysseus-dev#1230 shipped the adaptive auto-derivation but left `DEFAULT_HARD_MAX = 200_000` as a hardcoded module constant in src/context_budget.py. Admins on premium APIs with large context windows (kimi-k2 / minimax-m3 at 1M, etc.) can use their full window today only by setting `agent_input_token_budget` explicitly — which then takes them off the adaptive auto-path entirely. ## What this PR changes - src/settings.py: register `agent_input_token_hard_max` in DEFAULT_SETTINGS, default 200_000 (matches `DEFAULT_HARD_MAX`). Inline comment documents the no-op semantics in the explicit branch. - src/agent_loop.py: read the setting at the call site and pass it as the `hard_max` kwarg of `compute_input_token_budget`. Defensive parsing — missing / non-int / zero values fall back to `DEFAULT_HARD_MAX`, so a misconfig cannot silently zero the budget. - src/tool_implementations.py: three friendly aliases for `manage_settings`: - "hard max" -> agent_input_token_hard_max - "token budget cap" -> agent_input_token_hard_max - "input budget cap" -> agent_input_token_hard_max Plus the existing "token budget" -> agent_input_token_budget keeps a matching shorter alias "input budget". - tests/test_context_budget.py: 6 new tests on top of the existing 6: - hard_max raises the auto ceiling (1M ctx + raised cap -> 85% of ctx) - hard_max lowers the auto ceiling (128K ctx + 50K cap -> 50K) - hard_max has no effect on the explicit branch - DEFAULT_SETTINGS contains the new key - manage_settings aliases are registered - the live get_setting path returns the override value, and malformed values fall back per the agent_loop defensive parsing 12 passed in 0.04s. No changes to the pure helper signature or semantics; odysseus-dev#1230's behavior is the default when the new setting is unset. ## How it lets users drop the explicit override Before this PR, on a 1M-context model: agent_input_token_budget = 900_000 (explicit) -> 900K [user override] agent_input_token_budget = <unset> (auto) -> 200K [HARD_MAX] After this PR, same model: agent_input_token_budget = <unset> agent_input_token_hard_max = 900_000 -> min(1M * 0.85, 900K) = 850K [auto, no override needed] The explicit-override path keeps working unchanged for users who prefer it.

…ndow scaling (odysseus-dev#4122) * fix(agent): don't let a materialized default budget defeat context scaling odysseus-dev#1230 scales agent_input_token_budget to the model's context window unless the user explicitly set a budget, detected via is_setting_overridden(). But the settings-save path materializes every DEFAULT_SETTINGS key into settings.json (load_settings merges defaults; handlers persist the merged dict), so the persisted default 6000 reads as "overridden" and the budget code takes the min(6000, ctx) branch — silently re-capping long-context models at 6000 for anyone who has ever saved a setting. This reintroduces the exact regression odysseus-dev#1170/odysseus-dev#1230 set out to fix. Add is_setting_customized() (saved value != default) and gate the scaling on it instead of mere presence. A persisted default is not a user choice. is_setting_overridden has exactly one consumer (this budget path), so the change is contained. Tests cover the materialized-default regression, a deliberately-chosen budget still being honoured, and the absent-key case. * fix(agent): rework context-budget fix per review (odysseus-dev#4122) Address RaresKeY's review: P2 (explicitness): is_setting_customized treated a saved value equal to the default as "not explicit", which ALSO blocked a user from deliberately pinning the default budget. Reframe the default value itself as the AUTO sentinel — agent_input_token_budget == DEFAULT_BUDGET means "scale to the model's context window", any other value is an explicit cap. A materialized default still reads as auto (fixing the original regression), and any non-default value the user chooses is now honoured. Drop the now-unused is_setting_customized helper. P2 (fallback context): auto-scaling trusted get_context_length() even when it returned only the bare DEFAULT_CONTEXT fallback (no endpoint-reported / known window), over-allocating on self-hosted/proxy setups. Add get_context_length_known() (also returns whether the window was actually discovered); the budget block passes 0 when unknown so auto-scaling stays conservative instead of inflating to an unproven window. hard_max stays auto-only — a deliberate explicit budget wins (odysseus-dev#1190); kept that contract and answered the reviewer's question rather than silently reversing it. * test(agent): lock the materialized-default budget regression (review on odysseus-dev#4121) Per WGlynn's review on the issue: add an end-to-end regression that saves an UNRELATED setting (which makes the settings-save path materialize the budget default into settings.json) and asserts the budget still auto-scales rather than re-reading as an explicit 6000 cap — locking the exact reopening shut. To make the test bite the production decision (not just re-derive it), extract `budget_is_explicit()` into src/context_budget.py and use it from the agent loop. It keys off value-vs-default (the default is the auto sentinel), NOT settings presence — which is the whole point, since the save path materializes defaults. Note: after this PR's rework, is_setting_overridden has ZERO production callers, so the merged-dict materialization smell can't reach any setting through a presence check today (WGlynn's durability concern). * fix(agent): bind the budget context window to its own provenance (review odysseus-dev#4122) RaresKeY caught a correctness bug in the fallback-context guard: stream_agent_loop kept only the `known` flag from get_context_length_known() and budgeted off the passed-in `context_length`, which can come from a *different* lookup. Two failures: - local endpoints are re-queried, so the passed value can be a stale DEFAULT_CONTEXT fallback while the fresh probe proves the real (smaller) served context — we'd scale off the stale value; - callers that don't pass context_length (scheduled tasks, teacher escalation, skill test runs, bg_monitor) were capped at 6000 even when a long window is discoverable. Extract budget_context_for_model() which returns the freshly-probed window when known else 0, binding the flag to the value it proves; the agent loop uses it. Regression tests cover the stale-fallback, no-arg-caller, and probe-error paths. * docs(agent): fix stale budget comments + tighten to the contract (review odysseus-dev#4122) - settings.py: an explicit budget is clamped to the window only — hard_max is auto-only (odysseus-dev#1190); drop the incorrect "and to hard_max". - is_setting_overridden docstring: drop the stale "adaptive budgets" example; point value-sensitive callers at context_budget.budget_is_explicit. - Tighten the budget-block comments to the contract (default = auto sentinel, non-default = explicit cap, hard_max = auto-only ceiling). Comment/docstring-only; no behaviour change. * docs(agent): correct budget issue citations (odysseus-dev#1190 → merged odysseus-dev#1230/odysseus-dev#1273) The context-budget contract (auto-sentinel, explicit budgets honoured, hard_max auto-only) merged via odysseus-dev#1230 — odysseus-dev#1190 was the earlier, closed, superseded PR. Re-point the contract comments at odysseus-dev#1230 (the live source, already cited for the auto-sentinel two lines up in settings.py). The configurable hard_max setting (`agent_input_token_hard_max`) was a reviewer requirement first raised on odysseus-dev#1190, omitted from the merged odysseus-dev#1230, and actually added in odysseus-dev#1273 — credit odysseus-dev#1273 for it and correct the test comment's history (it previously implied this PR completed the requirement). Comment/docstring-only; no behaviour change. ---------

…ndow scaling (odysseus-dev#4122) * fix(agent): don't let a materialized default budget defeat context scaling odysseus-dev#1230 scales agent_input_token_budget to the model's context window unless the user explicitly set a budget, detected via is_setting_overridden(). But the settings-save path materializes every DEFAULT_SETTINGS key into settings.json (load_settings merges defaults; handlers persist the merged dict), so the persisted default 6000 reads as "overridden" and the budget code takes the min(6000, ctx) branch — silently re-capping long-context models at 6000 for anyone who has ever saved a setting. This reintroduces the exact regression odysseus-dev#1170/odysseus-dev#1230 set out to fix. Add is_setting_customized() (saved value != default) and gate the scaling on it instead of mere presence. A persisted default is not a user choice. is_setting_overridden has exactly one consumer (this budget path), so the change is contained. Tests cover the materialized-default regression, a deliberately-chosen budget still being honoured, and the absent-key case. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(agent): rework context-budget fix per review (odysseus-dev#4122) Address RaresKeY's review: P2 (explicitness): is_setting_customized treated a saved value equal to the default as "not explicit", which ALSO blocked a user from deliberately pinning the default budget. Reframe the default value itself as the AUTO sentinel — agent_input_token_budget == DEFAULT_BUDGET means "scale to the model's context window", any other value is an explicit cap. A materialized default still reads as auto (fixing the original regression), and any non-default value the user chooses is now honoured. Drop the now-unused is_setting_customized helper. P2 (fallback context): auto-scaling trusted get_context_length() even when it returned only the bare DEFAULT_CONTEXT fallback (no endpoint-reported / known window), over-allocating on self-hosted/proxy setups. Add get_context_length_known() (also returns whether the window was actually discovered); the budget block passes 0 when unknown so auto-scaling stays conservative instead of inflating to an unproven window. hard_max stays auto-only — a deliberate explicit budget wins (odysseus-dev#1190); kept that contract and answered the reviewer's question rather than silently reversing it. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * test(agent): lock the materialized-default budget regression (review on odysseus-dev#4121) Per WGlynn's review on the issue: add an end-to-end regression that saves an UNRELATED setting (which makes the settings-save path materialize the budget default into settings.json) and asserts the budget still auto-scales rather than re-reading as an explicit 6000 cap — locking the exact reopening shut. To make the test bite the production decision (not just re-derive it), extract `budget_is_explicit()` into src/context_budget.py and use it from the agent loop. It keys off value-vs-default (the default is the auto sentinel), NOT settings presence — which is the whole point, since the save path materializes defaults. Note: after this PR's rework, is_setting_overridden has ZERO production callers, so the merged-dict materialization smell can't reach any setting through a presence check today (WGlynn's durability concern). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(agent): bind the budget context window to its own provenance (review odysseus-dev#4122) RaresKeY caught a correctness bug in the fallback-context guard: stream_agent_loop kept only the `known` flag from get_context_length_known() and budgeted off the passed-in `context_length`, which can come from a *different* lookup. Two failures: - local endpoints are re-queried, so the passed value can be a stale DEFAULT_CONTEXT fallback while the fresh probe proves the real (smaller) served context — we'd scale off the stale value; - callers that don't pass context_length (scheduled tasks, teacher escalation, skill test runs, bg_monitor) were capped at 6000 even when a long window is discoverable. Extract budget_context_for_model() which returns the freshly-probed window when known else 0, binding the flag to the value it proves; the agent loop uses it. Regression tests cover the stale-fallback, no-arg-caller, and probe-error paths. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(agent): fix stale budget comments + tighten to the contract (review odysseus-dev#4122) - settings.py: an explicit budget is clamped to the window only — hard_max is auto-only (odysseus-dev#1190); drop the incorrect "and to hard_max". - is_setting_overridden docstring: drop the stale "adaptive budgets" example; point value-sensitive callers at context_budget.budget_is_explicit. - Tighten the budget-block comments to the contract (default = auto sentinel, non-default = explicit cap, hard_max = auto-only ceiling). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(agent): correct budget issue citations (odysseus-dev#1190 → merged odysseus-dev#1230/odysseus-dev#1273) The context-budget contract (auto-sentinel, explicit budgets honoured, hard_max auto-only) merged via odysseus-dev#1230 — odysseus-dev#1190 was the earlier, closed, superseded PR. Re-point the contract comments at odysseus-dev#1230 (the live source, already cited for the auto-sentinel two lines up in settings.py). The configurable hard_max setting (`agent_input_token_hard_max`) was a reviewer requirement first raised on odysseus-dev#1190, omitted from the merged odysseus-dev#1230, and actually added in odysseus-dev#1273 — credit odysseus-dev#1273 for it and correct the test comment's history (it previously implied this PR completed the requirement). Comment/docstring-only; no behaviour change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

nickorlabs force-pushed the fix/adaptive-input-token-budget branch from 84bb2b2 to 0646490 Compare June 2, 2026 14:29

nickorlabs force-pushed the fix/adaptive-input-token-budget branch from 0646490 to ee57701 Compare June 2, 2026 14:58

pewdiepie-archdaemon closed this Jun 2, 2026

This was referenced Jun 2, 2026

Make context-budget hard_max configurable via agent_input_token_hard_max setting (carry-over from #1190 review) #1272

Closed

fix(agent): make context-budget hard_max configurable via agent_input_token_hard_max setting (closes #1272) #1273

Merged

nickorlabs deleted the fix/adaptive-input-token-budget branch June 2, 2026 16:51

nsgds mentioned this pull request Jun 13, 2026

fix(agent): don't let a materialized default budget defeat context-window scaling #4122

Merged

6 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(agent): adaptive default for agent_input_token_budget (closes #1170)#1190

fix(agent): adaptive default for agent_input_token_budget (closes #1170)#1190
nickorlabs wants to merge 1 commit into
odysseus-dev:mainfrom
nickorlabs:fix/adaptive-input-token-budget

nickorlabs commented Jun 2, 2026 •

edited

Loading

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

nickorlabs commented Jun 2, 2026

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

nickorlabs commented Jun 2, 2026

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

nickorlabs commented Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Semantics

What this PR changes

Backwards compatibility

Tests

Diff

Adjacent context

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

nickorlabs commented Jun 2, 2026

What changed

Configurable hard max

Three symbols now

Tests

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

nickorlabs commented Jun 2, 2026

What changed in this revision

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

pewdiepie-archdaemon commented Jun 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

nickorlabs commented Jun 2, 2026 •

edited

Loading