You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I noticed that after changing the execute_threshold_percentage configuration, Opencode shows an error toast saying I’ve exceeded the model’s context window.
To preface, I’m using gpt-5.3-codex, which has a 400k context window, and magic-context has already been shrinking the context over time. Even at 30%, that should still be around 120k tokens, so I wouldn’t expect this limit to be hit.
Afterwards, it starts compacting using the built-in Opencode compaction and not magic-context compaction.
Once I reverted the change to the configuration, it seemed to work just fine.
One more thing I noticed: in the TUI popup for /ctx-status the execute threshold stays at 65% and doesn't reflect the configuration.
One more thing I noticed: in the TUI popup for /ctx-status the execute threshold stays at 65% and doesn't reflect the configuration.
This one suggests that there was an error with the config which loaded the default values for everything (65% exec thresh) but that doesn't explain the Input exceeds context window model.
Were you able to reproduce? Also if your log still has logs for when that compaction happened, we can understand better.
this was the log at the bottom of the doctor output
also I can reproduce this every time the configuration is set
regarding this:
One more thing I noticed: in the TUI popup for /ctx-status the execute threshold stays at 65% and doesn't reflect the configuration.
I think it might not be reading the config correctly and displaying it because the compaction occurs when I set the configuration so it is properly reading it somewhere...
Okay found the issue: execute_threshold_percentage minimum configurable value is 35 and unfortunately when lower than 35 is set, currently, the whole plugin disables itself. This will be fixed in the next release. (also lowering the min to 20)
Activity
[-]Opencode Forces Compression[/-][+]Opencode Forces Compaction[/+]ualtinok commentedon Apr 9, 2026
Could you share the full config file and
bunx @cortexkit/opencode-magic-context@latest doctoroutput.
null-axiom commentedon Apr 9, 2026
config that failed (i removed the execute_threshold_percentage after and it was fine):
doctor output:
ualtinok commentedon Apr 9, 2026
Interesting,
This one suggests that there was an error with the config which loaded the default values for everything (65% exec thresh) but that doesn't explain the Input exceeds context window model.
Were you able to reproduce? Also if your log still has logs for when that compaction happened, we can understand better.
null-axiom commentedon Apr 9, 2026
magic-context.log
this was the log at the bottom of the doctor output
also I can reproduce this every time the configuration is set
regarding this:
I think it might not be reading the config correctly and displaying it because the compaction occurs when I set the configuration so it is properly reading it somewhere...
ualtinok commentedon Apr 9, 2026
Okay found the issue: execute_threshold_percentage minimum configurable value is 35 and unfortunately when lower than 35 is set, currently, the whole plugin disables itself. This will be fixed in the next release. (also lowering the min to 20)
null-axiom commentedon Apr 9, 2026
I know this isn't the right thread for it but can u look at #3 again...
I left a comment on it, but since it's closed I'm not sure you'll see it again.
fix: defeat Bun static analysis of transformers import in Desktop
fix: close council audit findings on budget derivation
fix(embedding): bump @huggingface/transformers to 4.1.0 and add cross…
fix(cache-gates): shape consistency and retry-safe marker cleanup
fix(pi-plugin): close council audit findings (#1, #3, #4, #5, #7) + P…
fix(plugin,scripts): apply council audit post-ship follow-up fixes
fix(pi): round-5 control-flow + identity fixes (HIGH #3/#4/C5 + MEDs …
24 remaining items
fix: round-3 council findings (seq=0 m0 gap, dreamer archived underco…
fix(audit): per-session map leak, FTS phantom-UNIQUE dedup, column gu…
fix(audit): full-codebase council round — 9 real fixes + false-positi…
fix(cache): resolve transform/cache-stability council audit (5 real b…
fix(security): resolve 2 ship-blockers + 2 mediums from v0.22.5 relea…
fix(ctx-reduce): round-2 audit batch 2 — Pi config bleed, sqlite proj…
fix: release-audit batch — node:sqlite CI coverage, embedding-runtime…
chore: add audit-trail comments and extract magic number (cortexkit#4,c…
fix: scope dreamer decay by project, session leak, XML escaping, leas…
fix: address round 2 audit findings — 10 fixes
refactor: close remaining audit findings (cortexkit#2,cortexkit#3,cor…
fix: config recovery, lower threshold min, unbundle transformers, err…
fix(parity): hunt #4 — ctx_expand output rendering parity (tool-pair,…
fix(parity): revert hunt #4's ctx_expand ordinal-domain unification —…
fix(parity): hunt #4 briefs closed — ctx_search TS result shape on mo…