Intelligence and reviews
Putnami Intelligence gives an entitled workspace hosted code navigation, shared audit attestations, and subscription reviews of immutable pull-request heads.
putnami cloud setup installs the local MCP and audit wiring. Availability is
still discovered from the linked workspace's subscription on every hosted
call; a local flag cannot grant entitlement.
Measure agent readiness
putnami agent-readiness measures how far coding agents can work alone in a git
repository. It needs no workspace, no sign-in and no cloud setup. The fastest
way to run it installs the CLI and the extension first:
curl -fsSL "https://putnami.dev/install.sh?run=agent-readiness" | bashWith the CLI already installed, pin the extension once for your user, then run the command from any repository:
putnami extensions install --user @putnami/intelligence
putnami agent-readinessputnami intelligence agent-readiness is the same command. It reads the git
history, prints four lines, and exits:
Reading 2,431 commits, 11 areas… done (48 s)
This codebase supports supervised agents (L2). 63% of recent changes land where an agent cannot yet work alone.
Sent: 14 KB — counts, area names and paths, no file contents. Inspect: putnami intelligence agent-readiness --print-payload
Report: https://app.putnami.xyz/r/8f3k2qThe command sends counts, area names and repository-relative paths only: no
file contents, no author names or emails, and no absolute paths. It runs
read-only git commands and leaves git status unchanged. It never prompts and
never retries a failed send. The report link opens the report after you sign
in.
The first account that opens a report link owns the report. Every report you own is listed at app.putnami.xyz/r, grouped by repository, with the level change between runs. The list names a repository by its languages, never by its name, because the name never leaves your machine. Delete a report from that list.
The agent readiness method lists every marker, its thresholds, and the command that reproduces it.
| Flag | Effect |
|---|---|
--print-payload |
Print the exact bytes the command would send, and send nothing. Author hashes and the collection time change on every run. |
--timeout <seconds> |
Stop after this many seconds (default 120, at most 3600). |
--json |
Print the result as a JSON envelope instead of the four lines. |
| Exit code | Meaning |
|---|---|
0 |
The payload was sent, or --print-payload printed it. |
1 |
The run timed out, or the command refused a payload that could expose private data. |
2 |
Usage error: an invalid flag, or the directory is not inside a git repository with at least one commit. |
4 |
No report came back: the service was unreachable, timed out, refused the payload, or failed. The message says whether anything was sent. |
Check review readiness
Use the workspace-scoped doctor before launching a review:
putnami cloud intelligence review doctor --workspace <workspace-id>Doctor and review status need intelligence.review.read. Review mutations need
intelligence.review.manage; commands that create or revoke worker keys also
need workspace.manage.
putnami cloud intelligence status --workspace <workspace-id> reads the
durable index ledger with a global operator session. It is a platform-operator
command, not the ordinary workspace readiness check.
Launch and inspect a review
A review is tied to an exact PR head. Pass the immutable head SHA and record an audit reason:
putnami cloud intelligence review launch \
--workspace <workspace-id> \
--pr <number> \
--head <sha> \
--reason "review before merge"
putnami cloud intelligence review status \
--workspace <workspace-id> \
--review-key <review-key>Use --issues <n,n> on launch to attach explicit issue numbers. Pause, resume,
and reconcile are workspace operations and require --reason. Cancellation
also requires the immutable review key and --confirm:
putnami cloud intelligence review cancel \
--workspace <workspace-id> \
--review-key <review-key> \
--reason "superseded head" \
--confirmRun a workspace worker
Self-hosted review execution is an advanced path. worker run enrolls an
ephemeral key, passes it privately to the separate review worker, and revokes
the key when the process exits:
putnami cloud intelligence review worker run \
--workspace <workspace-id> \
--once \
--pull--pull fetches only absent images pinned by digest. Use --engine codex|claude-code to select the review engine.
Owner credentials are imported from an independently obtained official subscription artifact over standard input:
putnami cloud intelligence review auth login codex --stdinThe CLI does not launch vendor OAuth. Use auth status to inspect custody and
auth replace or auth revoke with the expected generation for concurrent
change protection.