senior-assess

Prompt file imported from Nom-nom-hub/senior-agent (.claude/commands/senior-assess.md). Fill in {{arguments}} before use. Copyright stays with the author.

Arguments

{{arguments}}

The input is an idea description (any size) and optionally slug=<slug>.

What this stage is

Ideas are cheap; misinvested engineering is not. This stage gathers evidence about an idea before a work slug exists — codebase fit, prior art, blast radius, unknowns — and issues exactly one verdict: go (worth building), needs-clarification (only named open questions stand between you and a go), or kill (with a written reason that stands without you). Killing an idea cleanly is a success, not a failure of ambition. Output: .senior/assessments/<slug>/assessment.md. No code is written.

Execution steps

1. Slug resolution: explicit slug= wins; else derive 2–4 kebab-case words from the idea. If .senior/assessments/<slug>/ exists, ask for a new slug in interactive use; in automated use append -2, -3… Never overwrite an existing assessment.

2. Restate the idea honestly in one sentence, then decompose: what it claims, what it implies, what it silently assumes (mark each assumption [NEEDS CLARIFICATION] rather than silently building on it).

3. Gather evidence (read-only):

   • Fit: does anything in the codebase already do this, partially or adjacent? (search concepts, docs, .senior/backlog.md for parked related items)
   • Cost shape: sketch, in sessions (walk / t-shirt / bag), what building the smallest valuable version would take — backend / frontend / data / infra split.
   • Constraints: does .senior/principles.md or the dossier rule anything out (technology, security posture, simplicity rules)?
   • Users: who benefits, how do we know they want it (evidence, not assertion)?

4. Pressure-test: state the strongest argument AGAINST doing this idea. Steelman the kill case before the go case; a verdict that never heard the other side is not an assessment.

5. Decide on the evidence:

   • go — fit is real, cost is acceptable, no load-bearing unknowns.
   • needs-clarification — the idea stands only if named questions get answers; list the questions (each answerable short — like the decide stage's rules) that block the decision, and nothing else.
   • kill — no fit, no users, cheaper alternative exists, or it contradicts the principles. Write the reason plainly; record a D-NNN decision in .senior/decisions.md so a future conversation doesn't re-litigate it.

6. Write assessment.md into .senior/assessments/<slug>/:

   # Idea Assessment: <title>
   
   - **Assessed**: <date> | **Decision**: go | needs-clarification | kill
   
   ## The idea (restated + decomposition)
   
   <one sentence + implied/silent assumptions>
   
   ## Evidence
   
   - Fit: <prior art / adjacent work / `[NEEDS CLARIFICATION: …]` markers>
   - Cost: <smallest valuable version + size + area split>
   - Constraints: <principles/dossier influences>
   - Users: <who + what proportion + why>
   
   ## The case against (steelmanned)
   
   <strongest kill argument>
   
   ## Open questions blocking the verdict
   
   <numbered, each short-answerable; state which resolution flips the decision>
   
   ## Next step
   
   - go → run `/senior-recon slug=<slug>` to start real work
   - needs-clarification → answer the listed questions, then re-assess
   - kill → stopped, decision recorded in .senior/decisions.md (D-NNN)

7. Report: the verdict, the strongest opposing argument you heard, and next step.

Rules

• Read-only on source; writes only inside .senior/assessments/<slug>/ (+ the D-NNN append for kill decisions).
• Spending hours to save days is the point — but an honest needs-clarification after focused evidence beats exhaustive research theater.
• The user's enthusiasm is evidence of DEMAND only if users other than them exist; name that distinction explicitly when relevant.

Done When

• assessment.md exists with exactly one decision, evidence-cited
• The case-against section steelmans the kill argument
• needs-clarification verdicts list only decision-blocking questions
• kill verdicts have a recorded D-NNN in .senior/decisions.md
• User told the verdict + next step + strongest opposing argument