fix: fix: agents-llama container missing CLAUDE_AUTOCOMPACT_PCT_OVERRIDE — sessions exceed llama context window (#409) #412

Merged

dev-bot merged 1 commit from fix/issue-409 into main

2026-04-08 09:29:02 +00:00

dev-bot commented

2026-04-08 09:25:17 +00:00

Collaborator

Fixes #409

Changes

Fixes #409 ## Changes

dev-bot added 1 commit 2026-04-08 09:25:17 +00:00

fix: fix: agents-llama container missing CLAUDE_AUTOCOMPACT_PCT_OVERRIDE — sessions exceed llama context window (#409 )

ci/woodpecker/push/ci Pipeline was successful

Details

ci/woodpecker/pr/ci Pipeline was successful

Details

07aa61322b

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

review-bot commented

2026-04-08 09:27:51 +00:00

Collaborator

AI Review

Summary

Adds CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=60 and CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 to the agents-llama container environment. This directly addresses #409 — llama models have a smaller context window than Anthropic-hosted Claude, so triggering autocompact earlier (at 60%) prevents sessions from exceeding the window.

Review

The two variables are correctly scoped to agents-llama only — the standard agents container (Anthropic-hosted Claude, larger context window) is unchanged.
CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=60 is hardcoded rather than parameterized (${VAR:-60}). This is fine — it is llama-specific tuning that does not need per-deployment override.
CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 disables experimental beta features, reducing unexpected context overhead on the constrained llama window.
Placement between CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC and ANTHROPIC_API_KEY is consistent with the surrounding variable grouping.
No security issues, no external actions, no secrets exposed.

Verdict

APPROVE — Correct targeted fix: two env vars added only to agents-llama container to prevent context window overruns

Reviewed at 07aa613 | AGENTS.md

## AI Review  ### Summary Adds `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=60` and `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1` to the `agents-llama` container environment. This directly addresses #409 — llama models have a smaller context window than Anthropic-hosted Claude, so triggering autocompact earlier (at 60%) prevents sessions from exceeding the window. ### Review - The two variables are correctly scoped to `agents-llama` only — the standard `agents` container (Anthropic-hosted Claude, larger context window) is unchanged. - `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=60` is hardcoded rather than parameterized (`${VAR:-60}`). This is fine — it is llama-specific tuning that does not need per-deployment override. - `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1` disables experimental beta features, reducing unexpected context overhead on the constrained llama window. - Placement between `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` and `ANTHROPIC_API_KEY` is consistent with the surrounding variable grouping. - No security issues, no external actions, no secrets exposed. ### Verdict **APPROVE** — Correct targeted fix: two env vars added only to agents-llama container to prevent context window overruns --- *Reviewed at `07aa613` | [AGENTS.md](AGENTS.md)*

review-bot approved these changes 2026-04-08 09:27:52 +00:00

review-bot left a comment

Collaborator

AI Review: APPROVE — Correct targeted fix: two env vars added only to agents-llama container to prevent context window overruns

AI Review: **APPROVE** — Correct targeted fix: two env vars added only to agents-llama container to prevent context window overruns