rudi193-cmd/skills/fleet-learning

Skill: fleet-learning

Purpose: Log fleet generation failures and fixes to fleet_learnings.jsonl using prompt_builder.log_learning() instead of silently patching and moving on.

Rules

1. WHEN

After running a fleet generation script (via llm_router.ask) and then manually fixing the output.

2. DO

Call prompt_builder.log_learning(task, provider, failure, fix) before moving on to the next step.

3. WHERE

Log file location:

C:/Users/Sean/Documents/GitHub/aios-minimal/prompts/fleet_learnings.jsonl

4. HOW

Import prompt_builder by adding aios-minimal to sys.path first:

import sys
sys.path.insert(0, "C:/Users/Sean/Documents/GitHub/aios-minimal")
from prompts import prompt_builder

5. ALSO

If the fix reveals a new constraint, update the prompt template in prompts/python_module.txt.

Minimal Code Example

import sys
sys.path.insert(0, "C:/Users/Sean/Documents/GitHub/aios-minimal")
from prompts import prompt_builder

task = "Generate fleet for async HTTP client"
provider = "openai/gpt-4"
failure = "Output missing timeout handling"
fix = "Added timeout parameter to request wrapper"

prompt_builder.log_learning(task, provider, failure, fix)

Why This Matters

Each unlogged fix is lost knowledge. When failures go unrecorded, the same workarounds are reinvented repeatedly, wasting tokens on repetitive troubleshooting. The fleet_learnings.jsonl log feeds future prompts with hard-earned lessons, enabling Claude to avoid known pitfalls and generate more robust fleet code from the start. Logging transforms one-off patches into reusable intelligence.