Target Loop

Target Loop Skill

Purpose

A structured iterative improvement loop that continues until a target artifact reaches a defined quality bar. Prevents premature stopping and systematic drift from the goal.

Step 1: Define the target

Before iterating, clearly define:

1. What is the artifact? (code, copy, design, document)
2. What is "done"? (specific quality criteria, not subjective)
3. What is the completion signal? (user says "good", tests pass, score threshold, etc.)
4. Maximum iterations? (default: 10)

Example completion criteria:

• "Copy: no filler words, every sentence earns its place, reads in under 60 seconds"
• "Code: all tests pass, no TypeScript errors, no console.logs"
• "Design: passes QA checklist, no overlapping elements, all text fits"

Step 2: Baseline assessment

Before iterating, assess the current state:

• What's working well? (preserve this)
• What's the biggest problem? (fix this first)
• Score: X/10 with specific reasoning

Step 3: Iterate

For each iteration:

ITERATION N/[MAX]
Current score: X/10
Biggest issue: [specific problem]
Fix applied: [what was changed]
New score: X/10
Remaining issues: [list]

Iteration priorities:

1. Fix the thing that matters most first
2. Don't fix multiple independent problems at once (hard to verify)
3. After each fix, re-evaluate the full artifact - one fix often creates another problem

Step 4: Completion check

Before stopping, verify:

• [ ] Does the artifact meet ALL stated completion criteria?
• [ ] Is there anything that would make a skeptical reviewer reject this?
• [ ] Has the core goal been preserved through iterations?

If any check fails, continue iterating.

Step 5: Final output

When done:

TARGET LOOP COMPLETE
Iterations: N
Final score: X/10
What changed: [summary of improvements]
Result: [the final artifact]

Anti-patterns

• Stopping at 8/10: "Good enough" is not the goal. If the bar was set, hit it.
• Fixing symptoms: If the same type of issue keeps appearing, there's a root cause to fix.
• Losing the core: Iterations can drift from the original goal. Re-read the target criteria every 3 iterations.
• Infinite loops: Set a max iteration count. If not converging, the criteria may need refinement.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: floomhq
• Source: floomhq/moto
• License: MIT
• Homepage: https://github.com/floomhq/moto#readme

Install and usage instructions live in the source repository linked above.