tokens&
BrowseFor enterprises
Grow product adoption
tokens&

Find tools, check provider offers, save a build plan, and share your work when you’re ready.

For buildersFor enterprises

For builders

  • Product launches
  • Builder projects
  • Startup credits and perks
  • Agent Skills
  • Published agents
  • API and agent docs
  • Publish a project

For enterprises

  • Start free company workspace
  • Submit a tool, product, or perk

Community

  • Community
  • Newsletter
  • Events
Xin

© 2026 tokensand, LLC. All rights reserved.

  • Terms
  • Privacy
  • Security
  • Data Processing
  • Status
  1. Home
  2. Projects
  3. Local review queue
Giovanni Amentaabout 9 hours agoVerified Engineer

Local review queue

A Python workflow that previews a local task, asks for approval, and recovers from failed or repeated runs without duplicate queue entries.

Demo

Live

Stack

0 tools

Open demoRemix this buildGive feedback
Launch note

Built

Local review queue

Proof

demo / build notes

Next

Try it, remix the project, or leave feedback.

Giovanni Amenta published this project so visitors can inspect the work and leave feedback.

Build Notes
Build prompt Problem: An operator needs repeatable tasks without duplicate work or unreviewed changes. First build: Create one local-only CLI workflow using the existing runtime and dependencies, with a read-only tool and an explicit review step. No new dependencies. Proof: Replay one task and simulate a failed tool call; show no duplicate action and a clear recovery. Publish proof: Link a working repo or demo in your project; choose public sharing when ready. Explore Google Agent Development Kit as a candidate tool; confirm its capabilities and account limits before choosing it. What I shipped Built from the Agent workflow for repeatable tasks challenge. This deterministic Python example uses only the standard library: one bounded read-only note tool, an explicit review step, and a SQLite transaction for the approved task. Try the source and instructions linked below. Preview a task, approve its exact digest, then rerun the command. The second run reports already_enqueued and retains one row. Remove the input file to simulate a tool failure, restore it, and retry. Verified on 16 November 2026: all 8 tests passed, including stale approval, changed input, concurrent retries, and process interruption before commit. A fresh-process replay left exactly one task. This is a Tokens& team-built learning example. It makes no model calls or network requests. Local database history is required for deduplication; filesystem power loss and hostile same-user changes are outside the tests. Repo and demo proof Source, README, and all 8 tests: https://gist.github.com/giolux1235/9dd9528a97ea6e0ab26ed931936372b1 Run python3 -m unittest -v in the downloaded source directory. Verified missing-file failure, recovery, approved enqueue, and replay: exactly one stored task. Tools used - What worked / what I learned A stable task ID plus one SQLite transaction makes retries safe within the retained local queue. Approval is tied to the reviewed content and destination; changed content needs a new review. No external side effects or provider calls are part of this example.
Feedback
Sign in to give feedback
Builder

Giovanni Amenta

Builder on tokens&

Public proof

Demo check

Reachable · HTTP 200

Publish your build