Workflow Elicitation
Capture operator judgment before you automate it.
Install
One-line install
npx attrition-sh pack install workflow-elicitation
AGENTS.md snippet (Claude Code / Cursor)
# Workflow Elicitation # See: /packs/workflow-elicitation
Raw Markdown
Machine-readable body for agent ingestion or copy/paste.
Telemetry
Not yet measuredSummary
A future-facing pattern for teams that need to externalize tacit expertise into reusable operating instructions before they try to scale an agent across a domain.
Fit and expected payoff
When this pack earns its extra structure, when to skip it, and what it should improve.
Use when
Situations where this pack earns its extra structure.
- The team has experienced operators whose judgment is hard to explain.
- Prompts keep getting longer because the real operating system is still implicit.
- You need a reusable role charter, escalation rules, and preferred answer shape.
Avoid when
Keeps the pack from becoming a default hammer.
- The workflow is already explicit and encoded in deterministic rules.
- You only need a lightweight prototype with minimal domain nuance.
What it improves
Expected outcomes if implemented well.
- Tacit knowledge is converted into reusable instructions and decision boundaries.
- The runtime gets better because the operating system is clearer.
- Model choice becomes less important than task specification quality.
Minimal instructions
Smallest useful starting point.
Do not start with the assistant. Start with an elicitation pass that captures: - the role charter - recurring decisions - escalation boundaries - trusted sources - preferred response structure Treat that operating system as an input to the runtime harness.
Full instructions
Complete natural-language instruction set.
When the workflow depends on expert judgment, add an elicitation stage before building the main assistant. The elicitation stage should capture: - what the operator is trying to optimize for - what facts must be cited - what decisions can be taken autonomously - what requires escalation - what sources outrank others - what a successful answer looks like Persist those outputs in a structured operating-system layer. Then feed that layer into the runtime harness, evaluation rubric, and role-specific UI.
Evaluation checklist
These checks should pass before you consider the pattern production-ready.
- Can the operating system be read without talking to the original expert?
- Are escalation rules and trusted sources explicit?
- Do runtime instructions reference this operating system instead of ad hoc prompt additions?
Common failure modes
Every check below traces back to a specific production failure. Read as: "I would think about X because in production Y can happen."
- Mid
The team keeps patching prompts instead of capturing the missing workflow rules.
- Trigger
- (legacy — trigger not separated)
- Prevention
- (legacy — no explicit prevention)
- Mid
Expert operators agree on outcomes but disagree on undocumented intermediate judgment.
- Trigger
- (legacy — trigger not separated)
- Prevention
- (legacy — no explicit prevention)
- Mid
The operating system is freeform notes with no reusable structure.
- Trigger
- (legacy — trigger not separated)
- Prevention
- (legacy — no explicit prevention)