Analyzing workflow patterns

Analyzing Workflow Patterns

Turn source material into evidence-backed integration choices. Understand the workflow before recommending changes and stop before every mutation.

Core protocol

1. Acquire every named source. Fetch a URL before interpreting it; identify the primary source and report anything unavailable.
2. Classify the material and extract actors, nouns, verbs, states, transitions, inputs, outputs, approval gates, failure paths, and reusable variables. Keep the abstract pattern separate from example identifiers.
3. Explain the workflow in the user's requested language and give a small and a scaled example. Default to the conversation language when none is requested.
4. Normalize the source into concise keyword chains so candidate flows use the same vocabulary and can be compared fairly.
5. Inspect the actual repository before designing options. Scan capabilities, not fixed provider paths:
   • shared and scoped instruction mechanisms;
   • tracker and planning conventions;
   • build, test, CI, release, and rollback commands;
   • agent, skill, handoff, and automation surfaces;
   • existing artifacts, ownership, and clean-canvas gaps.
6. Compare source needs with repository evidence. Mark every inference as [ASSUMPTION] and distinguish present, absent, conflicting, and unknown capabilities.
7. Generate three flows for a bounded change, four for a cross-cutting change, or five for a high-risk or organization-wide change. Each flow must include prerequisites, exact artifacts, approval gates, validation, risk, rollback, and repo-evidence rationale using flow-definition.md.
8. Score every flow with the all-higher-is-better model in scoring.md and render fit-matrix.md. State sensitivity or uncertainty when close scores depend on assumptions.
9. Recommend a flow or an explicit combination, explain the decisive trade-off, and stop. Do not create files, tracker objects, branches, commits, or remote resources until the user chooses a flow and separately approves its mutation plan.

Optional profiles

Load a profile only when the user requests it or repository evidence requires it. Profiles may add constraints but may not remove the core evidence, approval, validation, rollback, or hard-stop rules.

• hungarian.md changes the explanation layer.
• github-issues.md adds issue-decomposition output.
• legacy-five-flow.md preserves the source collection's exact eleven-phase, five-variant workflow.

Use template-variables.md to resolve generic tokens. Load output-quality.md for difficult reviews. The legacy phases.md is loaded only by the legacy-five profile.

Optional iteration contract

After the decision checkpoint, offer additions or questions only when they can materially change fit, safety, cost, or validation. Keep them specific and make clear that answering is optional. Do not append a fixed boilerplate list.