Failure mode to avoid
The fastest way to lose momentum is jumping from edit to handoff without checks. CATBot's loop keeps verification explicit so small mistakes do not cascade.
Execution Workflow
A deterministic loop from intent to verified output.
CATBot follows a clear delivery cadence that keeps actions visible, results testable, and handoff quality high across both quick tasks and larger feature work.
Stages
Every stage has a concrete output.
-
01
Frame the objective
Capture success criteria, constraints, and known risks in explicit language.
-
02
Collect repository context
Inspect source files, configs, and runtime assumptions before planning edits.
-
03
Design a scoped change set
Break work into minimal updates to keep review and rollback straightforward.
-
04
Execute with visibility
Apply edits while maintaining clear progress updates and command traceability.
-
05
Validate outcomes
Run available checks, compare behavior, and close correctness gaps quickly.
-
06
Deliver handoff summary
Report changed files, rationale, and logical next actions for the team.
Typical request rhythm
Use this structure for repeatable engineering execution across tasks.
- Start with a clearly bounded objective and target files.
- Use fast reads and search to map existing behavior.
- Make focused edits that align with repository patterns.
- Run checks where possible and note any gaps explicitly.
- Provide concise, actionable delivery notes.
delivery-loop.ps1
$goal = Define-Goal
$context = Read-Context
$plan = Scope-Changes
Apply-Edits $plan
Run-Checks
Write-Summary
Expected Artifacts
What should exist by the end of each loop.
Context Notes
A concise model of relevant code and assumptions for the current task.
Change Set
File-level edits that map directly to requested outcomes.
Validation Record
Checks performed, results observed, and residual risk if checks are unavailable.
Next
Connect the workflow to CATBot runtime layers and guardrails.
Architecture explains exactly where planning, tools, and policy intersect.