Agent Decision

Purpose

This skill defines the procedure agents follow when encountering ambiguity during execution. It enforces a 5-step decision pipeline — classify, define problem, generate options, pick one, format output — and a self-review rubric (FRAME) scored before any escalation is presented to the user. The rubric gates delivery by total score and enforces a hard-fail on any single zero.

Procedure

1. Classify the ambiguity. When you encounter missing or unclear information, determine which of these 5 types applies:

   • Approach choice — multiple valid approaches exist and the user did not specify one, but all lead to the same outcome. Example: whether to use a helper function or inline the logic.
   • Minor ambiguity — the intent is clear but a detail is vague and a reasonable default exists. Example: "add logging" without specifying log level — default to info.
   • Ambiguous requirement — the user's intent could mean two or more meaningfully different things. Example: "make it faster" could mean optimize the algorithm, add caching, or reduce payload size.
   • Missing information — a required fact is absent and cannot be inferred. Example: "deploy to the server" with no server specified and no convention to fall back on.
   • Risk confirmation — the action is destructive, expensive, or irreversible and the user has not explicitly authorized it. Example: dropping a database table, force-pushing to main.

2. Apply the branch.

   • Approach choice and minor ambiguity → proceed with a documented default. Record your choice in your handoff. Do not block.
   • Ambiguous requirement → escalate using the 1-3-1 method (step 3) and output template (step 4).
   • Missing information or risk confirmation → stop and ask directly. For non-interactive sessions, return a handoff explaining the gap.
   • The dividing line: if the ambiguity changes what you build, escalate. If it only changes how you build the same thing, proceed.

3. 1-3-1 analysis. When escalating, build your escalation in this order:

   • 1 problem — Write one sentence stating the single problem.
   • 3 options — Generate three meaningfully different approaches. Each with a one-sentence trade-off.
   • 1 recommendation — Pick the strongest option. Write the reason.

4. Apply the output template. Format the escalation exactly as follows:

   The [problem statement in one sentence].
   
   I'm [what I'm working on] and [where I am in the process].
   
   The choice is [decision in plain language]. No jargon.
   
   RECOMMENDATION: [A|B|C] because [reason].
   
   A) [option] — [one-sentence trade-off]
   B) [option] — [one-sentence trade-off]
   C) [option] — [one-sentence trade-off]

5. Score each criterion. After preparing the escalation, read the FRAME rubric below and assign a score of 0, 1, or 2 to each letter. Show the scoring breakdown to yourself (internal reasoning, not to the user).

6. Apply the hard-fail rule. If any letter scores 0, do not deliver — go to step 7 immediately.

7. Determine action by total score:

   • 9 – 10 — DELIVER — Escalation meets all criteria. Deliver to user.
   • 7 – 8 — FIX the scored < 2 criteria. a. Identify which letters scored below 2. b. Fix those gaps automatically (do NOT consult the user). c. Re-score, then deliver if 9-10. d. If still below 9, retry once more. e. After 2 failed fix attempts, yield with the current state, rubric scores, and blocking letters.
   • 0 – 6 — RESTART — The escalation is fundamentally broken. Discard and rebuild with corrected understanding, or yield with an explanation of what went wrong.

FRAME Rubric