Building an agentic workflow, part 1: deciding what done means

Description and use

The author outlines building an offline dot‑finance app and designing an agentic development loop with specification, context rules, correction limits, and independent review.

Problem

Manual reconciliation of credit‑card spending across spreadsheets is painful and error‑prone.

Outcome

Defined a clear, documented agentic workflow with explicit rules for context persistence, correction limits, and review, preparing for controller implementation.

Workflow

• Define application requirements in spec-file.md
• Create development workflow in development-workflow.md
• Document agent operating rules in AGENTS.md
• Use Codex to review spec and generate GitHub issues via to-issues skill
• Implement sequential issue loop: select ready issue, branch, run checks, review, integrate
• Set correction limits (3 attempts or 90 minutes) and record state in GitHub and local checkpoint
• Handle interruptions and platform failures by preserving correction count and pending checks
• Include architecture diagrams (LikeC4) and enforce update in PRs
• Plan prototype to establish desktop and LikeC4 build foundation

Evidence

I still reconcile my credit‑card spending manually across several spreadsheets. The calculations and automation I've added help, but it's still painful to do every month.

Source

I want an application that imports transactions, works out how much needs to come from different accounts, and records what's been settled.

Source

I began with a specification and asked Codex to review it and grill me on anything unclear before planning the work.

Source

The initial loop is sequential. It selects one ready issue with completed dependencies, works on an isolated branch, runs the required checks, and sends the result through independent review before integration.

Source

For the prototype phase, I've set an initial limit of three correction attempts or 90 minutes of active agent work, whichever comes first.

Source