New ideas that arrive mid-task land here as one dated line each. They are not expanded, researched, or built until the active gate passes and they are pulled deliberately.
- 2026-07-30: Send a finished report to someone by email. Note when pulling this: a report embeds quoted source code, so emailing it sends the reviewed code somewhere, which the README currently says does not happen, and SMTP means storing a credential, which SECURITY.md says the app never does. Consider the no-credential route first (export the report and hand it to the user's own mail client) with real SMTP as a clearly labelled opt-in.
Kept so the record of what was captured, and where it went, survives. Removing the line entirely would lose the trail.
- 2026-07-31 Reasoning effort is per review (low/medium/high/max, default
high). BUILT in commit f91c208, the same commit that filed this line, so it
was a note rather than an idea. The fact about the CLI accepting exactly
four levels lives where it is needed, in the comment at
src/lib/engine/command.ts. - 2026-07-31 The new-review screen needs a text area for the author description and a usage panel showing cached versus fresh tokens. BUILT in commit 935145a. One detail in the original line was wrong: the usage panel belongs on the review screen, not the new-review screen, because there is nothing to count before a review has run.
- 2026-08-04: retag unmigrated-consumer from rule 4 (index signature, which the Prefs interface does not have) to a rule that names the class, possibly a new rule 15.
- 2026-08-04: a risk-tagged variant of the quality gate, so the risk-first ordering is exercised with non-empty tags somewhere outside the pipeline tests.
- 2026-08-04: S3 ruleCode arrives from real models as prose ("Rule 11") where the ruleset speaks bare codes ("11"); normalise or validate at the stage boundary (observed in both clean fable runs, D-59).