feat(agent): Ask When Risky — a reviewer model rates each command before it runs by ndemianc · Pull Request #112 · levelcodeai/levelcode

feat(agent): Ask When Risky — a reviewer model rates each command before it runs

ndemianc

committed

46f943b

test(agent): approvals-editor-check.js runs Ask When Risky in the real editor

ndemianc

committed

d3d0441

docs: Ask When Risky — how a command is decided, what the reviewer trusts, and how it differs from ChatGPT's

ndemianc

committed

0c8ec05

fix(agent): the reviewer works on the native Anthropic route

ndemianc

committed

05060c1

fix(agent): a reviewer answer without an injection rating is unreadable, not "no injection"

ndemianc

committed

0907225

fix(agent): a command too long for the reviewer to see whole is never reviewed

ndemianc

committed

5f3f368

fix(agent): every run of a command is reviewed — an earlier pass is never reused

ndemianc

committed

23a2d98

fix(agent): only a goal the person typed is their word to the reviewer

ndemianc

committed

d24d0b5

fix(agent): a change to the approvals mode in Settings or another window takes effect at once

ndemianc

committed

e9ff3cc

test(agent): the typed-goal rule runs in a test, not only as a regex

ndemianc

committed

46a790b

docs: Ask When Risky after review — whole commands only, all three ratings, every run reviewed, typed goals only

ndemianc

committed

8d4ebb4

Navigation Menu

Repository navigation

feat(agent): Ask When Risky — a reviewer model rates each command before it runs - #112

feat(agent): Ask When Risky — a reviewer model rates each command before it runs#112
ndemianc merged 11 commits into
developfrom
feat/agent-review-mode

Commits on Oct 19, 2026

feat(agent): Ask When Risky — a reviewer model rates each command before it runs

test(agent): approvals-editor-check.js runs Ask When Risky in the real editor

docs: Ask When Risky — how a command is decided, what the reviewer trusts, and how it differs from ChatGPT's

fix(agent): the reviewer works on the native Anthropic route

fix(agent): a reviewer answer without an injection rating is unreadable, not "no injection"

fix(agent): a command too long for the reviewer to see whole is never reviewed

fix(agent): every run of a command is reviewed — an earlier pass is never reused

fix(agent): only a goal the person typed is their word to the reviewer

fix(agent): a change to the approvals mode in Settings or another window takes effect at once

test(agent): the typed-goal rule runs in a test, not only as a regex

docs: Ask When Risky after review — whole commands only, all three ratings, every run reviewed, typed goals only