Clarify managed human command checkpoint stability (#61800)

Document managed `prompt-user` command checkpoints as stable in the interactive
Codex CLI, while keeping app-server integration marked experimental. Clarify
that each matching command requires a fresh human decision, even with automatic
review enabled, and that session or persistent approvals cannot satisfy it.

GitOrigin-RevId: 98c36d00294bd97ab937f32940dfb24310ecbc32
This commit is contained in:
felixxia-oai
2026-11-16 12:05:55 +00:00
committed by copyberry
parent 0370e34fe8
commit 4fc37acc87
+16 -12
View File
@@ -1,19 +1,23 @@
# Managed human command checkpoints (experimental)
# Managed human command checkpoints
A matching managed `requirements.toml` prefix rule with `decision = "prompt-user"`
requires a fresh human decision, including when the thread uses automatic review.
Prefix matching, exact-path precedence, and executable mappings stay unchanged. Clients
supporting this workflow must initialize with `experimentalApi: true` and
`requiredHumanApproval: true`.
Managed `requirements.toml` prefix rules with `decision = "prompt-user"` are stable
in the interactive Codex CLI. A matching command requires a fresh human decision
before it can run, including when automatic review is enabled. Session and persistent
approvals cannot satisfy the checkpoint. Rules follow the usual command prefix matching
semantics, including exact-path precedence and executable mappings.
## App-server integration (experimental)
App-server clients supporting this workflow must initialize with `experimentalApi: true`
and `requiredHumanApproval: true`.
The server sends `item/commandExecution/requestApproval` with `kind: "humanCommand"`,
a unique `approvalId`, and the command to display. Only `accept`, `decline`, and
`cancel` are valid responses. Session and persistent approvals cannot satisfy the
checkpoint. Only capable clients receive or answer these requests. Pending prompts
use the normal approval replay flow when a capable client reconnects. With no capable
connection when approval is requested, the command is rejected. The prompt itself
does not start an execution item; normal command execution events report execution
after approval.
`cancel` are valid responses. Only capable clients receive or answer these requests.
Pending prompts use the normal approval replay flow when a capable client reconnects.
With no capable connection when approval is requested, the command is rejected.
Once an approved command starts, normal command execution events report its
execution item.
# Thread memory summary (experimental)