review

/anty:review — 5-Question Review Engine

/anty:review — 5-Question Review Engine

Trigger

User runs /anty:review

Behavior

Execute the 5-question review engine using the Analyst subagent (agents/analyst.md). Evaluate in strict priority order, proceeding deeper only if the answer to each question is "yes". Surface the level where the problem is.

Agent

Use the Analyst subagent for all evaluation steps.

The 5 Questions (Strict Priority Order)

Question 1: WHERE — Has the Goal progressed since last review?

Read .anty/goals/*.yaml and .anty/history/scans/*.yaml.

  • Compare current goal metric to last review's metric
  • Calculate absolute and percentage change
  • Check if progress rate is sufficient to reach target within timeline (including buffer)

Display:

Q1: Has the Goal progressed since last review?
----------------------------------------------------
Goal: [goal name]
Last review: [metric value] ([date])
Current:     [metric value] ([date])
Change:      [+/- delta] ([+/- %])
On track:    [yes/no — based on required run rate]
----------------------------------------------------

If NO (goal has not progressed): STOP HERE. The problem is at the WHERE level.

  • Diagnosis: Goal is not moving. No further investigation needed until progress resumes.
  • Recommend: Review which Drivers should be generating movement.
  • Skip Q2-Q5.

If YES: Proceed to Q2.

Question 2: Execution — Were approved Actions executed as planned?

Read .anty/actions/*.yaml — check actions with status approved or in_progress.

  • Compare planned steps vs completed steps for each action
  • Check if execution timeline matches planned timeline
  • Identify any actions that were approved but not started
  • Identify partially completed actions

Display:

Q2: Were approved Actions executed as planned?
----------------------------------------------------
Action: [action name]
  Planned steps: [n] | Completed: [m] | Status: [status]
  Timeline: [on-time / delayed by N days]
...
Overall execution rate: [completed/planned] ([%])
----------------------------------------------------

If NO (actions not executed as planned): STOP HERE. The problem is at the Execution level.

  • Diagnosis: Strategy exists, actions were approved, but execution failed.
  • Recommend: Investigate execution blockers (capacity, skills, dependencies).
  • Skip Q3-Q5.

If YES: Proceed to Q3.

Question 3: Planning — Was the Action plan well-constructed?

Read completed actions and their outcomes from .anty/actions/*.yaml and .anty/scratchpad.md.

  • Check if action steps were logically sequenced
  • Check if resource estimates were accurate
  • Check if dependencies were identified
  • Check if the plan had clear success criteria
  • Compare expected outcomes vs actual outcomes

Display:

Q3: Was the Action plan well-constructed?
----------------------------------------------------
Action: [action name]
  Steps logical:     [yes/no]
  Estimates accurate: [yes/no — expected N, actual M]
  Dependencies met:  [yes/no]
  Success criteria:  [defined/undefined]
  Outcome vs plan:   [matched / partial / missed]
----------------------------------------------------

If NO (plan was poorly constructed): STOP HERE. The problem is at the Planning level.

  • Diagnosis: Right actions chosen, but plan construction was flawed.
  • Recommend: Improve planning process (better estimation, dependency mapping, success criteria).
  • Skip Q4-Q5.

If YES: Proceed to Q4.

Question 4: HOW — Were the chosen Actions appropriate?

Read .anty/drivers/*.yaml, .anty/actions/*.yaml, and .anty/learning/choice-history.yaml.

  • Check if actions targeted the Crux Driver (highest impact_weight)
  • Check if actions addressed the constraint (not non-constraints)
  • Check if action selection aligned with learned preferences
  • Check if alternative actions might have been more effective
  • Reference action templates and their success rates

Display:

Q4: Were the chosen Actions appropriate for the situation?
----------------------------------------------------
Crux Driver: [driver name] (weight: [%])
Actions targeted crux: [yes/no]
Constraint-focused:    [yes/no]
Template success rate: [%] (based on [n] similar executions)
Alternative analysis:  [description of potentially better options]
----------------------------------------------------

If NO (wrong actions chosen): STOP HERE. The problem is at the HOW level.

  • Diagnosis: Root cause was correct, but the response actions were wrong.
  • Recommend: Re-evaluate action selection, consider different approaches.
  • Skip Q5.

If YES: Proceed to Q5.

Question 5: WHY — Was the root-cause analysis correct?

Read .anty/drivers/*.yaml, .anty/goals/*.yaml, and recent scan history.

  • Check if the identified constraint is actually the binding constraint
  • Check if impact_weights reflect observed reality
  • Check if the causal model (Driver -> Goal) is validated by data
  • Look for evidence that a different driver might be the true constraint

Display:

Q5: Was the root-cause analysis correct?
----------------------------------------------------
Identified constraint: [driver name]
Evidence supporting:   [list of supporting data points]
Evidence against:      [list of contradicting data points]
Alternative hypotheses:
  - [alternative constraint 1]: [supporting evidence]
  - [alternative constraint 2]: [supporting evidence]
Confidence: [high/medium/low]
----------------------------------------------------

If NO: The problem is at the WHY level.

  • Diagnosis: The fundamental understanding of what drives the goal is wrong.
  • Recommend: Restart root-cause analysis. Run /anty:scan for fresh data.

If YES: All 5 levels check out — the strategy is sound and executing well. Continue current course.

Summary Output

After completing the review (stopping at the appropriate level):

====================================================
  REVIEW RESULTS
====================================================
Q1 Goal Progress:    [PASS/FAIL]
Q2 Execution:        [PASS/FAIL/SKIPPED]
Q3 Planning:         [PASS/FAIL/SKIPPED]
Q4 Action Selection: [PASS/FAIL/SKIPPED]
Q5 Root Cause:       [PASS/FAIL/SKIPPED]

Problem Level: [WHERE/Execution/Planning/HOW/WHY/None]
Diagnosis: [one-line diagnosis]
Recommendation: [one-line recommendation]
====================================================

Save Results

Append review results to .anty/scratchpad.md:

---
[REVIEW] 2026-03-20 14:00 — Problem at [level] level
- Q1: [PASS/FAIL] | Q2: [PASS/FAIL/SKIP] | Q3-Q5: [status]
- Diagnosis: [summary]
- Recommendation: [summary]

Edge Cases

  • If no previous review exists, use goal creation date as baseline for Q1
  • If no actions exist, Q2 automatically fails (nothing to execute)
  • If no choice history exists, Q4 skips preference alignment check