Skip to content

fix(gradient): persist forceMinLayer and auto-recover from context overflow #21

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
BYK merged 2 commits into from
Feb 28, 2026
Merged

fix(gradient): persist forceMinLayer and auto-recover from context overflow #21

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cbef09f
fix(gradient): persist forceMinLayer and auto-recover from context ov…
BYK Feb 28, 2026
b04d705
chore: track pnpm-lock.yaml
BYK Feb 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 10 additions & 0 deletions AGENTS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,10 @@

### Gotcha

<!-- lore:019ca60f-0390-74ce-9381-b4296d94a553 -->
* **React useState async pitfall**: React useState setter is async — reading state immediately after setState returns stale value in dashboard components
<!-- lore:019ca60f-036f-7341-bbb7-dda2ea63294f -->
* **TypeScript strict mode caveat**: TypeScript strict null checks require explicit undefined handling
<!-- lore:019c91d6-04af-7334-8374-e8bbf14cb43d -->
* **Calibration used DB message count instead of transformed window count — caused layer 0 false passthrough**: Multiple gradient calibration bugs caused context overflow: (1) Calibration used DB message count instead of transformed window count — after compression (e.g. 50 msgs), delta saw ~1 new message → layer 0 passthrough → all raw messages sent → overflow. Fix: use getLastTransformedCount(). (2) actualInput formula omitted cache.write — on cold-cache turns actualInput became ~3 instead of 150K → layer 0 → overflow. Fix: include cache.write in both the formula and calibration guard. (3) Trailing pure-text assistant messages after tryFit cause Anthropic prefill errors. But messages with ANY tool parts must NOT be dropped (SDK converts to tool\_result user-role). Drop predicate: \`hasToolParts\`, not \`hasPendingTool\`. (4) Stats PATCH on message parts was removed — it was write-only dead code causing system-reminder persistence bug. Don't mutate parts you don't own.
<!-- lore:019c91c0-cdf3-71c9-be52-7f6441fb643e -->
Expand All @@ -36,11 +40,17 @@

### Pattern

<!-- lore:019ca60f-0371-7021-9a40-bf59215147e1 -->
* **Kubernetes deployment pattern**: Use helm charts for Kubernetes deployments with resource limits
<!-- lore:019c8ae9-2e54-7276-966a-befe699db589 -->
* **Use SDK internal client for HTTP requests in OpenCode plugins**: OpenCode plugins should use \`(ctx.client as any).\_client.patch()\` instead of raw fetch() with ctx.serverUrl. The \_client is the HeyAPI Client with correct base URL and interceptors, avoiding ConnectionRefused in TUI-only mode. Supports path interpolation. Caveat: \_client is private/undocumented. Note: lore's use of this pattern was removed — prefer dedicated endpoints over part mutation.

### Preference

<!-- lore:019ca60f-0376-72e0-805c-e1927dacdbed -->
* **General coding preference**: Prefer explicit error handling over silent failures
<!-- lore:019ca60f-0335-786c-83bd-a5cfcccbf223 -->
* **Code style**: User prefers no backwards-compat shims, fix callers directly
<!-- lore:019ca19d-fc02-7657-b2e9-7764658c01a5 -->
* **Code style**: User prefers no backwards-compat shims — fix callers directly. Prefer explicit error handling over silent failures. Derive thresholds from existing constants rather than hardcoding magic numbers (e.g., use \`raw.length <= COL\_COUNT\` instead of \`n < 10\_000\`). In CI, define shared env vars at workflow level, not per-job.

Expand Down
Loading