Fresh Eyes

Re-examine work as if seeing it for the first time. Familiarity breeds blind spots. This skill forces perspective shifts at three stages: before building, after building, and after fixing.

Core Philosophy

• Assume your first understanding is incomplete.
• Read code and plans literally, not through the lens of intent.
• Treat "obvious" as a warning sign, not a green light.
• Question decisions that feel automatic.
• Surface findings without filtering for convenience.

Modes

Planning Mode

Trigger: About to implement a feature, fix, or refactor. Plan exists but work has not started.

Purpose: Audit the plan for hidden complexity, missing acceptance criteria, and tech debt traps before committing to implementation.

Deliverables:

• Explicit acceptance criteria (extracted or written)
• Hidden complexity inventory
• Pre-mortem scenarios (what kills this in production?)
• Minimal test plan tied to acceptance criteria

See references/planning-prompts.md for full prompt templates.

Review Mode

Trigger: Code written and ready for review. You wrote it yourself or have been staring at it long enough to lose objectivity.

Purpose: Multi-pass review that forces you to see the diff as a stranger would. Each pass has a different lens.

Deliverables:

• Pass 1 findings: first impressions, naming confusion, flow surprises
• Pass 2 findings: edge cases, error paths, boundary conditions
• Pass 3 findings: regression risks, integration assumptions
• Structured findings (Blocker / Major / Minor / Structural Follow-up)

Repeat condition: Stop after 2 consecutive clean passes or a fixed timebox. Do not loop indefinitely.

See references/review-prompts.md for full prompt templates.

Reflection Mode

Trigger: Bug fixed or incident resolved. The obvious fix worked (or did not). You want to understand why it happened and whether it will happen again.

Purpose: Build causal chains from symptom to root cause. Distinguish root causes from symptoms. Extract tech debt signals and regression test guidance.

Deliverables:

• Causal chain (symptom to root cause, each link justified)
• Root vs symptom determination with evidence
• Tech debt signals (patterns that enabled the bug)
• Regression test recommendations

See references/reflection-prompts.md for full prompt templates.

Routing Table

Chaining with rick-rubin

fresh-eyes surfaces findings. rick-rubin provides scope-aware response to those findings.

Typical chain:

1. fresh-eyes Planning mode audits the plan.
2. rick-rubin (prompt A or B) trims scope based on findings.
3. Work proceeds with tighter scope and fewer blind spots.

Or after review:

1. fresh-eyes Review mode produces findings.
2. rick-rubin (prompt E or F) evaluates which findings require action vs deferral.

Do not duplicate rick-rubin's scope discipline logic. Cross-reference only.

Rationalizations (All Rejected)

Guardrails

• Do not use fresh-eyes as a delay tactic. Each mode has concrete deliverables and exit conditions.
• Do not loop Review passes indefinitely. Two clean passes or a timebox, then stop.
• Do not propose fixes in Planning mode. Surface problems only.
• Do not expand scope during Review. Flag structural issues as follow-ups, not blockers.
• For routine execution work, prefer lower-cost model tiers. Escalate for ambiguous root-cause reasoning. Reference model-router when explicit routing guidance is needed.