feat: add /cpo — product decision layer (kill criteria + outcome tracking)

[0x2kNJ]

Summary

• Adds /cpo — the product decision layer. Structured product decisions with mechanical gates, kill criteria, and a persistent decision journal.
• Same pattern as /cso — a named role that does one thing completely. /cso audits your attack surface. /cpo decides what to build, whether to build it, and tracks whether you were right.
• Fills the gap between /office-hours (idea validation) and /plan-eng-review (architecture lock-in). The only skill in the stack that can say "don't build this."
• Registered in proactive suggestions, generated for both Claude and Codex hosts.

Lifecycle position

THINKING              DECIDING              EXECUTING             FEEDBACK
─────────             ─────────             ─────────             ─────────
/office-hours    ──►  /cpo             ──►  /plan-eng-review  ──►  /retro
/plan-ceo-review ──►  (decides what    ──►  /build            ──►  /review
/retro           ──►   to build)       ──►  /ship             ──►  /qa
                                        ──►  /land-and-deploy  ──►  /canary

What it does

Three-phase gated flow: [FRAME] → [PATHS] → [VERDICT]

• Frame: Classifies one-way vs two-way door, identifies dominant Truth (user/strategic/economic/regulatory/execution), runs live market scan for irreversible bets, four premise checks
• Paths: Three structurally distinct options (never bold/balanced/conservative), recommendation first, pressure-test before committing
• Verdict: Kill criteria (metric + threshold + timeframe), confidence level, Truth fingerprint

Gates are mechanical — AskUserQuestion enforced, not suggestive.

The journal

Every verdict logs to ~/.cpo/decisions/*.yaml. CPO scans prior decisions at session start and surfaces related entries. --outcome closes the loop: replays the decision, walks through kill criteria with current data, surfaces patterns across all closed decisions.

Signal bus

• Reads: ~/.cpo/signals/{qa,review,retro,canary}-latest.yaml — surfaces red flags in [FRAME] before the decision is made
• Writes: ~/.cpo/signals/cpo-latest.yaml after every verdict — /build and /review can verify a decision exists before committing

Integration with existing skills

From	To	How
/office-hours	/cpo	--decide handoff when a strategic fork emerges
/plan-ceo-review	/cpo	Routes scope-level decisions
/cpo	/plan-eng-review	K) Eng brief saves artifact, suggests next
/cpo	/build, /ship	L) Hand off routes to execution
/retro	/cpo	Surfaces recurring patterns as decision input

Flags

--go (all-in-one), --quick (≤300 words), --deep (10-section), --journal, --review, --outcome, --save-context, --decide

Test plan

• bun run gen:skill-docs generates cpo/SKILL.md without errors
• bun run gen:skill-docs --host codex generates .agents/skills/gstack-cpo/SKILL.md
• bun test passes
• Manual: /cpo should we add a free tier? runs full FRAME→PATHS→VERDICT flow
• Manual: Decision journal entry written to ~/.cpo/decisions/
• Manual: --outcome replays a past decision and writes outcome block
• Manual: Proactive suggestion fires when user discusses a product decision

🤖 Generated with Claude Code