no-task-output
npx machina-cli add skill parcadei/Continuous-Claude-v3/no-task-output --openclawFiles (1)
SKILL.md
722 B
Never Use TaskOutput
TaskOutput floods the main context window with agent transcripts (70k+ tokens).
Rule
NEVER use TaskOutput tool. Use Task tool with synchronous mode instead.
Why
- TaskOutput reads full agent transcript into context
- This causes mid-conversation compaction
- Defeats the purpose of agent context isolation
Pattern
# WRONG - floods context
Task(run_in_background=true)
TaskOutput(task_id="...") // 70k tokens dumped
# RIGHT - isolated context, returns summary
Task(run_in_background=false) // Agent runs, returns summary
Source
- Session where TaskOutput caused context overflow
Source
git clone https://github.com/parcadei/Continuous-Claude-v3/blob/main/.claude/skills/no-task-output/SKILL.mdView on GitHub Overview
This skill prevents TaskOutput usage, which floods the main context with 70k+ tokens. It enforces using the Task tool in synchronous mode to maintain isolated, concise context and reliable summaries.
How This Skill Works
Rule: never call TaskOutput. Instead, run Task with run_in_background=false to execute synchronously and return a summary. This preserves context isolation, avoids mid-conversation overflow, and yields a concise result rather than a full transcript.
When to Use It
- When a session risk of context overflow due to long transcripts is high
- When you need isolated agent context and want to avoid cross-talk between tasks
- When a concise summary is preferred over the full transcript after a Task run
- When debugging or auditing conversations where transcripts could be excessively large
- When operating under strict context window limits or cost constraints
Quick Start
- Step 1: Identify any TaskOutput usage in your flows and remove it
- Step 2: Replace with Task(run_in_background=false) to execute synchronously and return a summary
- Step 3: Validate the summary contains the needed insights and stays within context limits
Best Practices
- Enforce the no-TaskOutput rule in code reviews and design guidelines
- Replace TaskOutput calls with Task(run_in_background=false) to obtain summaries
- Remove any remaining TaskOutput references from existing flows
- Document the pattern in team onboarding and development docs
- Test flows with long transcripts to ensure summaries stay within context limits
Example Use Cases
- A chatbot session previously flooded with TaskOutput now returns a concise summary via Task(run_in_background=false), preserving context
- A knowledge-base update uses Task to fetch results and returns a summary instead of full transcripts
- An orchestration pipeline replaces TaskOutput usage with synchronous Task calls to keep contexts isolated
- A debugging scenario demonstrates context overflow with TaskOutput and is corrected by adopting the summary pattern
- An NLP agent workflow runs tasks synchronously, ensuring transcript isolation and efficient context usage
Frequently Asked Questions
Add this skill to your agents
