Keep small task-specific starting files for coding agents
At a glance
A developer uses short Markdown files to carry selected context into fresh agent sessions. This workflow is listed under Coding.
- Agents / tools
- Coding agents
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Create separate Markdown starting files for specific domains or task types rather than a large general project description. The commenter usually keeps each file under thirty lines. 2) Record useful module or task preferences, pointers to references, and concrete choices such as preferring one existing pattern over another. At the start of a fresh agent session, explicitly point the agent to the relevant file. 3) When an implementation needs a correction because the agent missed a necessary convention, add that observed lesson to the appropriate file; the agent can help update it using information from the current task. 4) Keep human control over what enters the starting context and use a fresh session when exploring a different approach.
How it went
The commenter says the files reduce repeated explanations and help avoid the distraction and bias toward abandoned approaches that arose in long sessions or from broad automatic context retrieval.