Enforce a four-checkpoint workflow with mandatory state tracking and verifiable evidence across code and data tasks.
npx skills add https://github.com/wjy0705/taskFlow-skill --skill taskflow-wjy0705
Please help me install this Agent Skill. Skill: taskflow Source: https://github.com/wjy0705/taskFlow-skill/tree/main Command: npx skills add https://github.com/wjy0705/taskFlow-skill --skill taskflow-wjy0705
Task Flow enforces a disciplined, verifiable AI workflow that prevents hallucinations and wasted tokens by mandating explicit stage checks, state tracking, and graceful degradation.
Trigger Task Flow with /taskflow to start the four-checkpoint protocol and guide you through understanding, capability selection, evidence verification, and safe stop.
High-intent search queries and answers about installing and using this skill.
Preventing AI hallucinations requires enforcing an explicit four-checkpoint workflow with mandatory state tracking to verify problem understanding, capability selection, and evidence-based validation. This approach uses verifiable outputs like stdout, diffs, or runnable test scripts to ensure accuracy.
An evidence-based verification workflow requires verifiable evidence such as stdout, diffs, or screenshots to validate AI outputs. When direct command execution is unavailable, it automatically generates runnable test scripts for human validation to ensure environment-bound accuracy.
Implement state tracking by mandating the AI to output current stage markers like [当前阶段：...] at the start of each reply. This provides full traceability across understanding, capability selection, verification, and fallback stages throughout coding and analysis tasks.
Yes, a structured guardrail workflow applies cross-platform to coding, debugging, operations, and text analysis tasks. It enforces graceful degradation by generating verifiable test scripts for humans when direct environment commands cannot run.
Environment graceful degradation handles validation failures by automatically generating runnable test scripts for humans to execute. This ensures that evidence-based verification continues even when the AI cannot directly run commands in the target environment.
AI workflows lack traceability when they skip explicit state tracking checkpoints during execution. Enforcing a four-checkpoint protocol with mandatory stage outputs at the start of each reply restores traceability across understanding, capability selection, verification, and fallback phases.
wangke19Resume coding sessions from .ai-session/STATE.md checkpoints or start fresh.
tuantiensiuExecute verification commands and collect test, lint, and build evidence before completion claims.
matt-rileyRun exact verification commands and quote fresh output as evidence.
bugrabilgeDesign autonomous AI agents with tool use, memory, and multi-agent orchestration.
