fork-test
Flagged{"isSafe":false,"isSuspicious":true,"riskLevel":"medium","findings":[{"category":"prompt_injection","severity":"medium","description":"The skill includes 'Execute any requested tool calls', which could be exploited by crafted prompts to force tool execution beyond intended scope. Lacks safeguards such as input validation or whitelisting for tool usage.","evidence":"2. Execute any requested tool calls"}],"summary":"Potential prompt-injection risk due to the instruction to 'Execute any requested tool calls' in the skill description. No concrete malicious commands are present, but the pattern could enable unsafe behavior if deployed without safeguards. The rest of the content is a benign skeleton for testing context fork behavior."}
npx machina-cli add skill Geeksfino/openskills/fork-test --openclawFork Test Skill
This skill exists only for testing the context: fork pattern.
Purpose
Validates that:
- Forked contexts properly isolate intermediate outputs
- Tool calls are recorded in the session
- Only summaries are returned to the parent context
- Multiple tool calls are tracked correctly
Instructions
When this skill is activated, perform the following actions:
- Read the provided input
- Execute any requested tool calls
- Return a concise summary of the results
Return the input unchanged as the result.
Source
git clone https://github.com/Geeksfino/openskills/blob/main/examples/skills/fork-test/SKILL.mdView on GitHub Overview
This skill tests the fork pattern by validating isolation of intermediate outputs in forked contexts, recording tool calls in the session, and ensuring only summaries reach the parent context. It is designed for testing, not production use.
How This Skill Works
When activated, the skill reads the input, executes any requested tool calls within the forked context, and then returns a concise summary of the results. The parent context receives only the summary, while intermediate results remain isolated in the fork. Finally, the skill returns the input unchanged as the result.
When to Use It
- When validating that forked contexts do not leak intermediate outputs between parent and fork
- When verifying that tool calls are recorded in the session for forked runs
- When ensuring only summaries propagate back to the parent context
- When testing multiple sequential tool calls within a fork
- When debugging forked-context behavior in a test suite
Quick Start
- Step 1: Read the provided input
- Step 2: Execute any requested tool calls
- Step 3: Return a concise summary of the results and the input unchanged as the result
Best Practices
- Run in non-production environments only
- Verify isolation of intermediate outputs within the fork
- Inspect session logs to confirm tool-call records
- Ensure the parent receives only summaries, not raw outputs
- Test multiple tool calls to validate tracking accuracy
Example Use Cases
- Unit test that a fork isolates intermediate data while logging tool calls
- Test that the parent only receives a summary after a forked run
- Verify the input is echoed unchanged in the forked context
- Debug a mismatch where summaries fail to reflect fork results
- Benchmark tool-call tracking limits within a single fork
