[morgaesis]

Adds a default total LLM request budget so hosted reviews stay inside the worker's outer watchdog even when no environment variable overrides it: 540 seconds total, reserving 120 seconds for the independent scorer and leaving a 60-second margin before the 600-second watchdog fires. Local and interactive runs are unaffected by default and keep no total budget unless POSTIL_LLM_TOTAL_TIMEOUT_SECS is explicitly set. Crate version bumped to 0.4.2, with new regression tests for the default budget and for a cascade retry expiring mid-request.

[postil-dev]

The default LLM total timeout is now always active, including for local runs, which may unexpectedly break long-running local reviews that previously had no total timeout.

• warn src/llm.rs:195 — Default total timeout now applies to local runs · confidence 0.90 · kind: humanEscalation

Model: deepseek/deepseek-v4-pro

[postil-dev]

The README incorrectly states that every review (local or hosted) is subject to a default 540-second total timeout, but the code applies that default only to hosted runs; local reviews have no total timeout unless the environment variable is set. This false claim could mislead users.

• error README.md:201 — README falsely claims default total timeout applies to local reviews · confidence 0.90 · kind: contentPolicy
• warn src/llm.rs:195 — Default total timeout now applies to local runs · confidence 0.90 · kind: humanEscalation

Model: deepseek/deepseek-v4-pro

[postil-dev]

The default per-request LLM timeout was reduced from 480s to 420s, which may cause timeouts for large diffs in local runs where no total budget is active. This change is not mentioned in the PR description and could be an unintended regression.

• warn src/llm.rs:144 — Default per-request LLM timeout reduced from 480s to 420s · confidence 0.70 · kind: humanEscalation
• error README.md:201 — README falsely claims default total timeout applies to local reviews · confidence 0.90 · kind: contentPolicy
• warn src/llm.rs:195 — Default total timeout now applies to local runs · confidence 0.90 · kind: humanEscalation

Model: deepseek/deepseek-v4-pro

[postil-dev]

The change reduces DEFAULT_REQUEST_TIMEOUT_SECS from 480 to 420 seconds for all execution modes. This may cause previously passing long-running local/CI reviews to time out. Confirm this is intentional.

• warn src/llm.rs:144 — Default request timeout quietly reduced from 480 to 420 · confidence 0.80 · kind: risk
• error README.md:201 — README falsely claims default total timeout applies to local reviews · confidence 0.90 · kind: contentPolicy
• warn src/llm.rs:195 — Default total timeout now applies to local runs · confidence 0.90 · kind: humanEscalation

Model: deepseek/deepseek-v4-pro

[postil-dev]

A new default LLM timeout budget is introduced for hosted remote reviews, with regression coverage added. The PR title and description contain a content-policy violation.

• error .postil/pr-description:2 — remove fabricated claim · confidence 0.25 · kind: uncertainty

3 finding(s) from the previous review resolved.

Model: mistralai/mistral-small-3.2-24b-instruct

[postil-dev]

A doc comment on HOSTED_LLM_REQUEST_TIMEOUT_SECS contains an arithmetic contradiction (claims 4*120=480s fits within 420s budget).

• error src/review.rs:32 — Doc comment contradicts constant value · confidence 0.95 · kind: contentPolicy

Model: deepseek/deepseek-v4-pro