Add policy-options skill

[beingkk]

What / why

New Claude Code skill (.claude/skills/policy-options/) that turns a Policy Atlas project's existing evidence base into a cited markdown of policy directions and interventions. Reads findings and sources via the local postgres-local MCP (read-only by transaction), writes to .outputs/policy-options/ (gitignored).

Not part of the task-cycle spine — this is a developer-facing tool, not a product capability. Small documentation touch adds a "Claude Code MCP (optional)" setup section to docs/agentic-ops/environment.md.

Task: n/a (tooling)

Proof it works

• make verify: not run — no production code changed.
• Manual: ran the skill end-to-end against the food-price-stabilisation project (27 IOFs, 69 ICFs, 10 sources). Output rendered with British English, selective bold, inline evidence citations and source URLs; verified locally.

Risk tier

Tier 0 — docs + a markdown-only skill. No schema, deps, auth, egress, or public interface changes. Nothing touches the running product.

AI role

Agent drafted the skill, environment doc addition and PR. Human directed scope, style rules, output location and read-only stance.

Review focus

Skill accuracy, evidence discipline, doc placement.

Reviews run

• /code-review (optional at Tier 0)
• Manual end-to-end run

Known gaps & deferred seams

• Options are not persisted back to the DB. Promote to a proper capability if reruns need to be tracked.
• No read-only DB role yet; transaction-level guard is the current control (documented).

Public safety

• No secrets, credentials, or real/acquired source text in the diff.
• Logs / traces / screenshots are public-safe.
• No approval-gated change snuck in unapproved.

🤖 Generated with Claude Code