From c742c22ed63efc79cac9c4e5bb6198c0fcade8e9 Mon Sep 17 00:00:00 2001 From: Tanisha Aberdeen <32620895+aliasunder@users.noreply.github.com> Date: Fri, 4 Sep 2026 20:19:41 -0400 Subject: [PATCH] chore: bump default context_budget_tokens from 80K to 300K --- README.md | 2 +- action.yml | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index a20ddcd..da1db49 100644 --- a/README.md +++ b/README.md @@ -86,7 +86,7 @@ The `@umm review` comment trigger lets you re-request a review on any PR by comm | `severity_threshold` | `low` | Minimum severity to post: `low` \| `medium` \| `high` \| `critical` | | `conventions_file` | `AGENTS.md` | Repo-relative path to the conventions file included in the prompt (truncated at ~8000 tokens). When the file also changed in the PR, deduplication ensures its full text appears exactly once across context channels | | `phases` | `combined` | How the review dimensions are dispatched: `combined` (one model call carrying every dimension), `parallel` (three focused calls at once — each reads deeper than the single call; wall clock is the slowest of the three and prompt tokens roughly triple), or `sequential` (the same three calls in order, each seeing the earlier findings). Findings two phases report on the same lines collapse to one, keeping the higher severity. Empty = `combined`, so workflows can wire an unset repo variable directly | -| `context_budget_tokens` | `80000` | Approximate token budget for prompt context (file contents + diff — conventions have a separate cap) | +| `context_budget_tokens` | `300000` | Approximate token budget for prompt context (file contents + diff — conventions have a separate cap) | | `trace_related_files` | `true` | Enable heuristic context scanning — import-tracing for caller regressions and mention-matching for doc staleness detection. Does not affect `priority_docs` | | `priority_docs` | `README.md` | Comma-separated repo-relative paths included in review context with a reserved 10% budget floor, independent of `trace_related_files`. Docs already present in full from other context channels are not re-read; a diff-only changed file is still read here so its full text reaches the model. Subject to the shared doc token budget (the floor prevents starvation by related files but does not guarantee inclusion when the budget is exhausted by the diff and changed files themselves). Empty = disabled | | `max_scan_files` | `5000` | Maximum files to walk during workspace scan for related file and doc detection | diff --git a/action.yml b/action.yml index bd697d7..2e4883e 100644 --- a/action.yml +++ b/action.yml @@ -43,7 +43,7 @@ inputs: context_budget_tokens: description: Approximate token budget for prompt context (file contents + diff — conventions have a separate cap) required: false - default: "80000" + default: "300000" trace_related_files: description: Include related files in the prompt — import-tracing for caller regressions, and doc-mention scanning for staleness detection required: false