What problem does it solve?
It automates the full lifecycle of a single queued engineering task in the Telegram Desktop C++ codebase, from task resolution and planning through implementation, multi-lens code review, evidence-based testing, and canonical commit publication, removing the manual coordination of these phases.
Core Features & Use Cases
- Single-task ownership: Resolves one task by slug or full id, claims it atomically, resumes interrupted or blocked work, and never processes other queue items.
- Adaptive review pipeline: Runs a mandatory general review plus five standard lenses (lifetime, reuse, structure, performance, security) with fast NOT_APPLICABLE bailouts and bounded convergence assessment.
- Falsifiable evidence loop: Designs and executes test campaigns using static checks, unit tests, Debug Telegram builds, in-app overlays, Computer Use, and screenshots with explicit oracles, controls, and falsifiers.
- Use Case: A maintainer invokes $perform-task with a task slug; the skill plans the change, implements it in the Telegram Desktop source, reviews the diff, proves behavior with runtime evidence, and publishes a canonical Approve commit.
Quick Start
Ask the assistant to run $perform-task with the short slug or full dated id of an existing ai-tdesktop task you want implemented and published.
