@lifeaitools/rdc-skills 0.35.8 → 0.35.9

package/commands/help.md

@@ -1,93 +0,0 @@
----
-name: help
-description: rdc:help () — list all MCP skills with usage, requirements and slash form
----
-
-> **Output contract:** Checklist/table output only. No raw MCP/JSON/log dumps.
-
-# rdc:help — Manifest-Driven Skill Reference
-
-Print a concise skill reference from the plugin manifest, then stop. Do not
-ask which command to run unless the user explicitly asks for a recommendation.
-
-## Source Of Truth
-
-Resolve the manifest in this order and use the first existing file:
-
-1. `{PLUGIN_ROOT}/.claude-plugin/plugin.json`
-2. `C:/Dev/rdc-skills/.claude-plugin/plugin.json`
-3. `~/.claude/plugins/cache/rdc-skills/rdc-skills/latest/.claude-plugin/plugin.json`
-
-Read `skills_meta` from the manifest. Accept either the current object map shape
-(`skills_meta.<name>`) or an older array shape (`skills_meta[]`); normalize to
-entries before rendering. Group entries by `category` in this order:
-`planning`, `build`, `deploy`, `release`, `dev-loop`, `reporting`, `tooling`,
-`infra`.
-
-For each row, show:
-
-```text
-{slash}  —  {usage}  —  needs: {requires or "(none)"}  —  cf: {yes|no}
-```
-
-At the end, render a short decision tree from `skills_meta[].triggers`.
-
-If the manifest is unreachable, say `manifest fallback used` and render the
-static fallback table from `skills/help/SKILL.md`.
-
-## Direct MCP / curl Access
-
-Show this block after the command table so non-Claude callers have the same
-entry point. Do not summarize, compress, relabel, or omit the curl examples;
-this block is the public direct-caller contract for MCP/API/curl users.
-
-```text
-MCP endpoint: https://rdc-skills.regendevcorp.com/mcp
-Health:       curl -s https://rdc-skills.regendevcorp.com/health
-Header:       Accept: application/json, text/event-stream
-Tools:        rdc_skill_list, rdc_skill_search, rdc_skill_get
-Variants:     cli for Claude Code/Codex/local terminal; cloud for claude.ai
-Response:     SSE `data:` line contains the JSON-RPC envelope; tool text is result.content[0].text
-Catalog:      rdc_skill_list returns aliases, args, requires, produces, follows, leads_to, output_contract, codeflow_required, and variants
-```
-
-Minimal curl examples:
-
-```bash
-curl -s -X POST https://rdc-skills.regendevcorp.com/mcp \
-  -H 'Content-Type: application/json' \
-  -H 'Accept: application/json, text/event-stream' \
-  -d '{"jsonrpc":"2.0","id":0,"method":"tools/list"}' \
-  | sed -n 's/^data: //p'
-
-curl -s -X POST https://rdc-skills.regendevcorp.com/mcp \
-  -H 'Content-Type: application/json' \
-  -H 'Accept: application/json, text/event-stream' \
-  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"rdc_skill_list","arguments":{}}}'
-
-curl -s -X POST https://rdc-skills.regendevcorp.com/mcp \
-  -H 'Content-Type: application/json' \
-  -H 'Accept: application/json, text/event-stream' \
-  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"rdc_skill_search","arguments":{"query":"turn this article into social posts"}}}'
-
-curl -s -X POST https://rdc-skills.regendevcorp.com/mcp \
-  -H 'Content-Type: application/json' \
-  -H 'Accept: application/json, text/event-stream' \
-  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"rdc_skill_get","arguments":{"name":"rdc:build","variant":"cli"}}}'
-
-curl -s -X POST https://rdc-skills.regendevcorp.com/mcp \
-  -H 'Content-Type: application/json' \
-  -H 'Accept: application/json, text/event-stream' \
-  -d '{"jsonrpc":"2.0","id":4,"method":"tools/call","params":{"name":"rdc_skill_get","arguments":{"name":"rdc:build","variant":"cli","format":"json"}}}'
-```
-
-Required exact token for tests and direct callers: `"format":"json"`
-
-## Hard Rules
-
-- `cf: yes` skills must consult CodeFlow before acting.
-- Never run `pnpm build` at monorepo root. Use scoped typecheck/tests.
-- Never commit to `main` without explicit approval.
-- Never overlap agents on the same files.
-- Update work items in real time for skills that require work-items RPCs.
-- Credentials come from clauth: `curl -s http://127.0.0.1:52437/v/<service>`.

package/commands/overnight.md

@@ -1,222 +0,0 @@
----
-name: overnight
-description: rdc:overnight ([scope]) — drain the work queue unattended, end to end
----
-
-> **⚠️ OUTPUT CONTRACT (READ FIRST):** `guides/output-contract.md`
-> Checklist-only output. No tool-call narration. No raw MCP/JSON/log dumps.
-> One checklist upfront, updated in place, shown again at end with a 1-line verdict.
-
-> If dispatching subagents or running as a subagent: read `{PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md` first (fallback: `{PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md`).
-
-> **Sandbox contract:** This skill honors `RDC_TEST=1` per `guides/agent-bootstrap.md` § RDC_TEST Sandbox Contract. Destructive external calls short-circuit under the flag.
-
-
-# rdc:overnight — Overnight Build Supervisor
-
-## When to Use
-- Starting an unattended multi-hour build session
-- Project lead says "run overnight", "build everything", "go while I sleep"
-- Kicking off a full epic queue after scoping is agreed
-
-## Arguments
-- `/rdc:overnight` — work all urgent/high priority todo epics
-- `/rdc:overnight <epic-id>` — work a specific epic only
-- `/rdc:overnight label=<label>` — work epics matching label
-
-## Phase 1 — Pre-flight
-
-**First action — set the overnight sentinel** so the `no-stop-open-epics` Stop hook engages (interactive sessions are not gated by it):
-
-```bash
-mkdir -p {PROJECT_ROOT}/.rdc && touch {PROJECT_ROOT}/.rdc/overnight.lock
-```
-
-**Last action at end of run (success OR failure) — remove the sentinel:**
-
-```bash
-rm -f {PROJECT_ROOT}/.rdc/overnight.lock
-```
-
-If the sentinel file does not exist, the Stop hook will NOT block — which means interactive sessions stop freely and only `rdc:overnight` is held to the "drain the queue" contract.
-
-Before touching any code, verify the environment is safe:
-
-1. **Clauth daemon alive:**
-   ```bash
-   curl -s http://127.0.0.1:52437/ping
-   ```
-   If not responding: report `BLOCKED: credential daemon offline` and exit. Do not proceed.
-
-2. **Git state clean:**
-   ```bash
-   git status --short
-   ```
-   Must be on the development branch with no uncommitted changes. If dirty: commit or stash first.
-
-3. **Baseline review:**
-   Run `rdc:review --unattended`. If `REVIEW_STATUS.verdict = "HAS_ISSUES"` and
-   issues cannot be auto-fixed: report `BLOCKED: codebase has pre-existing issues` and exit.
-   A dirty baseline overnight compounds into a disaster by morning.
-
-If all three pass: proceed to Phase 2.
-
-## Phase 2 — Load Scope
-
-Determine which epics to work:
-
-- **Specific epic arg:** load that one epic via `get_work_items_by_epic()`
-- **Label filter:** `SELECT get_open_epics(p_label_filter := '<label>')`
-- **No arg (default):** `SELECT get_open_epics()` filtered to:
-  - `priority IN ('urgent', 'high')`
-  - `status IN ('todo', 'in_progress')`
-  - Order: urgent first, then high, then by `created_at`
-
-Log the epic queue at the start of `.rdc/overnights/overnight-<YYYY-MM-DD>.md` (fallback: `.rdc/overnights/overnight-<YYYY-MM-DD>.md` if `.rdc/` does not exist).
-
-## Phase 3 — Epic Loop
-
-For each epic in the queue, run this sequence:
-
-### 3a. Research (if needed)
-Condition: epic has 0 child tasks AND no matching doc in `.rdc/plans/` (or `.rdc/plans/` as fallback)
-
-```
-rdc:preplan <topic> --unattended
-```
-
-Check `PREPLAN_STATUS.recommendation_confidence`:
-- `"high"` or `"medium"`: proceed
-- `"low"`: escalate via advisor tool (see Escalation Protocol below)
-  - If advisor gives direction: proceed with that direction
-  - If advisor cannot resolve: skip this epic, log reason, move to next
-
-### 3b. Plan (if no tasks exist)
-Condition: epic has 0 child tasks after preplan
-
-```
-rdc:plan <epic-id> --unattended
-```
-
-Check `PLAN_STATUS.task_count > 0` before continuing.
-If 0 tasks created: escalate via advisor, then skip if still unresolved.
-If `PLAN_STATUS.held_for_design_review > 0`: do not silently promote those tasks to todo. Log the held task IDs and their durable assessment, request human Design Review, and continue only with the independently admitted portion of the epic.
-
-### 3c. Build
-
-This skill delegates to rdc:build which uses typed agent dispatch. See rdc:build for agent type classification.
-
-```
-rdc:build <epic-id> --unattended
-```
-
-Agents receive the relevant guide file from `.rdc/guides/` (fallback: `.rdc/guides/`) based on their work package type.
-
-After each wave: check `BUILD_STATUS`. If `escalated: true`, log the escalation
-in the overnight doc and continue — don't stop the loop.
-After each wave and after every resumed epic, require the `runOrchestrator()` receipt from `rdc:build`. `admission_refocus` or `pipeline_blocked` means the epic is held for durable Design Review or validator closure; log that state and do not hand-reconstruct a dispatch wave.
-
-### 3d. Review
-
-```
-rdc:review --unattended
-```
-
-Check `REVIEW_STATUS.verdict`:
-- `"CLEAN"`: re-run the durable CodeFlow supervisor. Mark an epic `done` only when its receipt is `pipeline_complete` and every executable child is design-review admitted and validator-closed (`status = done`); otherwise preserve the epic/task hold and log `admission_refocus` or `pipeline_blocked`.
-- `"HAS_ISSUES"` with `escalations > 0`: log issues, push what's clean, continue
-- `"HAS_ISSUES"` with `escalations = 0` (all auto-fixed): push, continue
-
-### 3e. Commit checkpoint
-
-After each epic (pass or fail):
-```bash
-if [ "$RDC_TEST" != "1" ]; then
-  git push origin {development-branch}
-else
-  echo "[RDC_TEST] skipping git push origin {development-branch}"
-fi
-```
-
-This ensures every epic's work is saved regardless of what comes next.
-
-## Phase 4 — Exit
-
-After all epics are processed:
-
-1. Run `rdc:report --unattended`
-
-2. Write session summary to `.rdc/overnights/overnight-<YYYY-MM-DD>.md` (fallback: `.rdc/overnights/overnight-<YYYY-MM-DD>.md`):
-   ```markdown
-   # Overnight Session — YYYY-MM-DD
-
-   ## Started
-   <timestamp>
-
-   ## Epics Attempted
-   | Epic | Status | Tasks Done | Commits |
-
-   ## Epics Completed
-   <list with epic IDs>
-
-   ## Escalations
-   N advisor calls — details:
-   - <epic>: <what was escalated> → <advisor response>
-
-   ## Blockers Remaining
-   <any epics skipped or partially done>
-
-   ## Git Summary
-   - Total commits: N
-   - Push status: ✅
-
-   ## Completed
-   <timestamp>
-   ```
-
-3. Final push:
-   ```bash
-   if [ "$RDC_TEST" != "1" ]; then
-     git push origin {development-branch}
-   else
-     echo "[RDC_TEST] skipping final git push origin {development-branch}"
-   fi
-   ```
-
-## Escalation Protocol
-
-The advisor tool pairs this executor with a high-level model for high-stakes decisions.
-Use it when genuinely stuck — not for every small uncertainty.
-
-**Escalation triggers:**
-- `PREPLAN_STATUS.recommendation_confidence = "low"` (too many unknowns)
-- `PLAN_STATUS.task_count = 0` after planning (nothing actionable)
-- Build agent fails twice on the same task
-- `REVIEW_STATUS` has unfixable issues requiring architectural judgment
-- Credential daemon goes down mid-session (escalate before exiting)
-
-**How to escalate:**
-Provide the advisor with:
-1. What was attempted and what failed
-2. The error or ambiguity in detail
-3. Two most likely paths forward with tradeoffs
-4. Which epic/task is blocked
-
-**After advisor responds:**
-- Log the guidance in the overnight doc
-- Resume the loop from where it stopped
-- If advisor cannot resolve: mark task/epic `blocked`, log reason, skip to next
-
-**Max escalations:** 3 per epic. After 3, skip the epic and log.
-
-## Safety Rules
-
-- Branch: development branch always — NEVER touch main/production
-- NEVER run `pnpm build` — use `npx tsc --noEmit` for typecheck, vitest for tests only on modified packages
-- NEVER let agents overlap on the same files
-- Push after every epic, not just at the end
-- Update Supabase work items in real time throughout
-- **A clean review never overrides a durable Design Review hold, an incomplete checklist, or missing validator closure**
-- Max 2 hours per epic — if exceeded, skip and log `TIMEOUT`
-- If credential daemon goes down mid-session: write current state to overnight doc, push, exit gracefully
-- If git push fails: log the failure, attempt rebase, retry once — do not force push

package/commands/plan.md

@@ -1,178 +0,0 @@
----
-name: plan
-description: rdc:plan (topic) — produce architecture, decisions and an epic with tasks
----
-
-> **⚠️ OUTPUT CONTRACT (READ FIRST):** `guides/output-contract.md`
-> Checklist-only output. No tool-call narration. No raw MCP/JSON/log dumps.
-> One checklist upfront, updated in place, shown again at end with a 1-line verdict.
-
-> If dispatching subagents or running as a subagent: read `{PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md` first (fallback: `{PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md`).
-
-
-# rdc:plan — Architecture & Work Packages
-
-## When to Use
-- After `/rdc:preplan` produced research findings
-- Project lead gives architectural direction ("build X with Y approach")
-- An epic exists but needs breakdown into implementable tasks
-- Before any large build session
-- Called by `rdc:overnight` when an epic has no child tasks
-
-## Arguments
-- `rdc:plan <topic>` — interactive planning session
-- `rdc:plan <epic-id> --unattended` — silent mode for overnight builds
-
-## Procedure
-
-1. **Load source documents — MANDATORY before any planning decisions.**
-
-   **Step 1a — Always load these regardless of topic:**
-   ```
-   .claude/rules/infrastructure-contract.md   — hard deployment + registry rules
-   .claude/rules/work-items-rpc.md            — work item schema, RPC, status enums
-   .claude/rules/system-quick-links.md        — routing map to all system architecture docs
-   .claude/rules/version-numbering.md         — version bump rules for affected packages
-   docs/CODING-STANDARDS.md                   — SOLID/Clean-Architecture standard; names
-                                                 ATF's Test-Ladder and rdc-harness's
-                                                 proof-ledger pattern where either applies
-                                                 (regen-root; skip if absent)
-   ```
-
-   **Step 1b — Identify affected domains, then load the matching architecture doc:**
-
-   | Domain keywords in topic | Architecture doc to read |
-   |--------------------------|---------------------------|
-   | PRT, trust, capital, NAV, investor, land, DST | `docs/systems/prt/ARCHITECTURE.md` |
-   | CS 2.0, HAIL, PAL, virtue, quad-pixel, ontology, BPMN, cognitive | `docs/systems/cs2/ARCHITECTURE.md` |
-   | marketing, CRM, campaign, contact, outreach, RDC app | `docs/systems/rdc/ARCHITECTURE.md` |
-   | Claude workflow, skills, agents, dispatch, rdc:build | `docs/systems/claude-workflow/ARCHITECTURE.md` |
-   | Life AI, LIFEAI platform, life.ai | `docs/systems/lifeai/ARCHITECTURE.md` |
-   | media, R2, images, regen-media, MCP image | `docs/systems/media/ARCHITECTURE.md` |
-   | UI, component, brand, design token, shared, OG image | `docs/systems/shared/ARCHITECTURE.md` |
-
-   If the topic spans multiple domains: read ALL matching architecture docs before proceeding.
-   A plan that contradicts an existing architecture doc is invalid — load them first.
-
-   **Step 1c — Load domain-specific rules and context files:**
-
-   | Domain | Additional files to read |
-   |--------|---------------------------|
-   | CS 2.0 / any CS2 paradigm work | `.claude/rules/cs2-architecture-first.md` |
-   | Database, schema, migrations, RPC | `.claude/context/supabase-schema.md` |
-   | UI, components, brand, tokens | `.claude/context/design-system-global.md` |
-   | Deploy, infrastructure, DNS, SSL | `.claude/context/coolify-deployment.md` |
-   | Credentials, MCP, clauth, subagents | `.claude/context/clauth.md` |
-   | OG images, social meta, brand assets | `.claude/context/brand-gate.md` |
-   | Cross-platform, Cowork, subagent MCP | `.claude/context/platform-cross-ref.md` |
-   | MCP server development | `.claude/context/mcp-server-auth.md` |
-
-   **Step 1d — Load CLAUDE.md for every affected package:**
-   - Identify which packages in `packages/` will be created or modified
-   - Read `packages/<name>/CLAUDE.md` for each one that has one
-   - Mandatory: `packages/supabase/CLAUDE.md` if any DB work is involved
-   - Mandatory: `packages/ui/CLAUDE.md` if any UI work is involved
-   - Read `packages/<name>/package.json` to understand current exports and dependencies
-
-2. **Gather additional inputs:**
-   - Research doc from preplan (if exists): `.rdc/research/<topic>.md` (fallback: `.rdc/research/<topic>.md`)
-   - Project lead's architectural direction from conversation
-   - Existing Supabase epics: `SELECT get_open_epics()`
-   - Check `prototype_registry` for any existing prototypes on this topic:
-     ```sql
-     SELECT name, component, source_path, status FROM prototype_registry
-     WHERE status IN ('prototype', 'converting') ORDER BY created_at DESC;
-     ```
-   - Check `design_context` for prior design decisions:
-     ```sql
-     SELECT topic, context_type, summary FROM design_context
-     WHERE topic ILIKE '%<topic>%' ORDER BY created_at DESC;
-     ```
-
-3. **Read the codebase** — understand current state:
-   - What packages are affected?
-   - What types/interfaces already exist?
-   - What tests exist?
-   - What's the dependency graph?
-
-4. **Make design decisions** — for each major choice:
-   - State the decision clearly
-   - Document what was chosen and what was rejected
-   - Explain WHY (tradeoff rationale)
-   - Note consequences and reversibility
-   - **Verify the decision does not contradict any loaded architecture doc** — if it does, flag the conflict before proceeding
-
-5. **Define work packages** — break into agent-dispatchable units:
-   - Each work package = one agent assignment
-   - No file overlap between packages
-   - Each package has: scope, files to create/modify, test requirements
-   - Assign an agent type to each work package from the typed dispatch table in rdc:build
-   - Include the guide file path (from `.rdc/guides/`, fallback `.rdc/guides/`) in each work package description
-   - Include any relevant architecture doc, context file, or package CLAUDE.md the agent must read
-   - Estimate: small (1 agent, <500 LOC), medium (1 agent, 500-1500 LOC), large (needs splitting)
-   - Attach a **Design Review contract** to every executable package. It must contain:
-     - architecture alignment: registered architecture-evidence reference(s), declared target boundary, and a concrete alignment claim;
-     - completeness: independently observable acceptance criteria plus at least one required `decomp-*` and `test-*` checklist row; and
-     - proportionality: estimated files/LOC, declared surfaces, and change kind.
-   - Do not invent architecture evidence. Missing, unregistered, broad-refactor, multi-surface, or disproportionate contracts are intentionally routed to human Design Review.
-   - **ATF Test-Ladder / rdc-harness (WIP — best-effort, not a hard gate yet):** if the
-     package being planned has its own ATF `STP-001.md` or ships an `rdc-harness`-style
-     `tools/proof-ledger.mjs`/`tools/mutate-check.mjs` pair, name it in the work package's
-     test requirements as an additional validation signal alongside vitest/tsc. Neither
-     system is fleet-wide or hook-enforced yet — do not fabricate a ladder or a proof
-     ledger for a package that doesn't already have one.
-
-6. **Write plan doc** to `.rdc/plans/<topic-slug>.md` (fallback: `.rdc/plans/<topic-slug>.md` if `.rdc/` does not exist):
-   ```markdown
-   # Plan: <Topic>
-   > Generated: <date> | Epic: <id if exists>
-
-   ## Source Documents Read
-   (list every architecture doc, rules file, context file, and package CLAUDE.md loaded in Step 1)
-
-   ## Goal
-   ## Design Decisions
-   ## Work Packages
-   (each package must include: agent type, guide file, architecture docs agent must read, files to create/modify, test requirements)
-   ## Sequencing (what can parallelize, what depends on what)
-   ## Risks & Mitigations
-   ## Architecture Doc Conflicts (if any)
-   ```
-
-7. **Create Supabase epic + child tasks:**
-   - Epic via `insert_work_item(p_item_type := 'epic', ...)`
-   - Immediately call `set_epic_governance_refs(p_epic_id, p_plan_ref := '.rdc/plans/<topic-slug>.md', p_spec_ref := '.rdc/plans/<topic-slug>.md', p_architecture_ref := '<ARCHITECTURE.md path or NULL — set only when this plan crosses an architectural boundary>', p_scoping_statement := '<in/out of scope, one paragraph>')` so `rdc:build`/`rdc:overnight`/`rdc:fixit`/`rdc:refactor` never have to guess or stall on these fields
-   - One task per work package only through `upsert_admitted_work_item(...)`, never a direct `work_items` write. Supply a stable source fingerprint, the package checklist, and the Design Review contract from Step 5.
-   - Read the durable result for each package:
-     - `dispatchable: true` / `automatic_approved` → task may become `todo`;
-     - `dispatchable: false` / `needs_human` or `pending` → keep the task `blocked`, label it `needs-human-design-review`, and include the assessment in the plan status.
-   - If (and only if) the epic's `architecture_ref` is set, each task's checklist passed to `upsert_admitted_work_item` MUST also include one required `architecture-fidelity-<slug>` row naming the specific architecture doc + boundary — the exit gate hard-rejects `done` on any task under an `architecture_ref` epic missing this row. Do not add it under an epic with no `architecture_ref` — that holds ordinary work for a review it doesn't need.
-   - Set priorities: urgent/high/normal based on sequencing
-
-8. **Report results:**
-   - Interactive: present the plan for approval before building
-   - Unattended: skip approval, proceed immediately, emit status block:
-     ```
-     PLAN_STATUS: { epic_id, task_count, dispatchable_task_count, held_for_design_review, doc_path, waves, source_docs_read: [list], architecture_conflicts: [] }
-     ```
-
-## Unattended Escalation
-
-When `--unattended` and genuine architectural ambiguity is detected — meaning multiple
-valid approaches exist with significantly different tradeoffs (not just minor style choices)
-— escalate via the advisor tool. Provide: the decision point, the options with tradeoffs,
-and the project context. Resume with advisor's recommendation. If advisor is unavailable,
-choose the most conservative/reversible approach and document the decision.
-
-## Rules
-- **Source documents in Step 1 are MANDATORY — a plan that hasn't read the architecture docs is invalid**
-- Interactive: ALWAYS get approval before proceeding to build
-- Unattended: proceed immediately without approval
-- Plan doc goes in `.rdc/plans/` (fallback: `.rdc/plans/` if `.rdc/` does not exist) — not `.planning/`
-- Each work package must be independently executable by an agent
-- No file overlap between work packages
-- Include test requirements in every work package
-- **No executable work item is dispatchable without a durable Design Review decision; a planner cannot self-approve by supplying prose-only evidence**
-- Reference affected CLAUDE.md files and architecture docs in each work package description
-- Reference the relevant guide file from `.rdc/guides/` (fallback: `.rdc/guides/`) for agent context
-- Always list source docs read in the output doc header and status block

package/commands/preplan.md

@@ -1,134 +0,0 @@
----
-name: preplan
-description: rdc:preplan (topic) — research the unknowns before committing to an architecture
----
-
-> **⚠️ OUTPUT CONTRACT (READ FIRST):** `guides/output-contract.md`
-> Checklist-only output. No tool-call narration. No raw MCP/JSON/log dumps.
-> One checklist upfront, updated in place, shown again at end with a 1-line verdict.
-
-> If dispatching subagents or running as a subagent: read `{PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md` first (fallback: `{PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md`).
-
-
-# rdc:preplan — Research Before Planning
-
-## When to Use
-- Starting a new feature area you haven't built before
-- Need to understand how best-in-class projects solve a problem
-- Codebase has unknowns that need mapping before planning
-- Project lead says "research", "look into", "what's the best way to", "how do others do"
-- Called by `rdc:overnight` before planning an epic with no existing tasks
-
-## Arguments
-- `rdc:preplan <topic>` — interactive research session
-- `rdc:preplan <topic> --unattended` — silent mode for overnight builds
-
-## Procedure
-
-1. **Parse the topic** from user input or epic title/description.
-   - Interactive: if vague, ask ONE clarifying question before proceeding
-   - Unattended: infer from the epic title + description — never pause to ask
-
-2. **Load source documents — MANDATORY before any analysis.**
-
-   **Step 2a — Always load these regardless of topic:**
-   ```
-   .claude/rules/infrastructure-contract.md   — hard deployment + registry rules
-   .claude/rules/work-items-rpc.md            — work item schema and RPC patterns
-   .claude/rules/system-quick-links.md        — routing map to system architecture docs
-   docs/CODING-STANDARDS.md                   — SOLID/Clean-Architecture standard; names
-                                                 ATF's Test-Ladder and rdc-harness's
-                                                 proof-ledger pattern where either applies
-                                                 (regen-root; skip if absent)
-   ```
-
-   **Step 2b — Identify affected domains, then load the matching architecture doc:**
-
-   | Domain keywords in topic | Architecture doc to read |
-   |--------------------------|---------------------------|
-   | PRT, trust, capital, NAV, investor, land, DST | `docs/systems/prt/ARCHITECTURE.md` |
-   | CS 2.0, HAIL, PAL, virtue, quad-pixel, ontology, BPMN, cognitive | `docs/systems/cs2/ARCHITECTURE.md` |
-   | marketing, CRM, campaign, contact, outreach, RDC app | `docs/systems/rdc/ARCHITECTURE.md` |
-   | Claude workflow, skills, agents, dispatch, rdc:build | `docs/systems/claude-workflow/ARCHITECTURE.md` |
-   | Life AI, LIFEAI platform, life.ai | `docs/systems/lifeai/ARCHITECTURE.md` |
-   | media, R2, images, regen-media, MCP image | `docs/systems/media/ARCHITECTURE.md` |
-   | UI, component, brand, design token, shared, OG image | `docs/systems/shared/ARCHITECTURE.md` |
-
-   If topic spans multiple domains, read ALL matching architecture docs.
-   If unsure which domain applies, read `docs/systems/claude-workflow/ARCHITECTURE.md` as the fallback.
-
-   **Step 2c — Load domain-specific rules and context files:**
-
-   | Domain | Additional files to read |
-   |--------|---------------------------|
-   | CS 2.0 / any CS2 paradigm work | `.claude/rules/cs2-architecture-first.md` |
-   | Database, schema, migrations, RPC | `.claude/context/supabase-schema.md` |
-   | UI, components, brand, tokens | `.claude/context/design-system-global.md` |
-   | Deploy, infrastructure, DNS, SSL | `.claude/context/coolify-deployment.md` |
-   | Credentials, MCP, clauth, subagents | `.claude/context/clauth.md` |
-   | OG images, social meta, brand assets | `.claude/context/brand-gate.md` |
-   | Cross-platform, Cowork, subagent MCP | `.claude/context/platform-cross-ref.md` |
-
-   **Step 2d — Load CLAUDE.md for every affected package:**
-   - Identify which packages in `packages/` are relevant to the topic
-   - Read `packages/<name>/CLAUDE.md` for each one
-   - At minimum read `packages/supabase/CLAUDE.md` if any DB work is involved
-   - At minimum read `packages/ui/CLAUDE.md` if any UI work is involved
-
-3. **Web research** — search for current (2025-2026) best practices:
-   - How do major projects solve this?
-   - What tools/libraries exist?
-   - What are the common tradeoffs?
-
-4. **Codebase analysis** — what do we already have?
-   - Search relevant packages for existing code
-   - Check `.rdc/research/` for prior research on this topic (fallback: `.rdc/research/`)
-   - Check `docs/archive/` for historical work
-   - Research agents should read relevant guides from `.rdc/guides/` (fallback: `.rdc/guides/`)
-   - Check work items for related epics
-
-5. **Best-in-class comparison** — create a comparison table:
-   | Approach | Pros | Cons | Fit for Us |
-
-6. **Surface unknowns** — what questions remain unanswered?
-
-7. **Write research doc** to `.rdc/research/<topic-slug>.md` (fallback: `.rdc/research/<topic-slug>.md` if `.rdc/` does not exist):
-   ```markdown
-   # Research: <Topic>
-   > Generated: <date> | Requested by: Project Lead
-
-   ## Source Documents Read
-   (list every architecture doc, rules file, context file, and package CLAUDE.md loaded in Step 2)
-
-   ## Question
-   ## What We Already Have
-   ## Best-in-Class Analysis
-   ## Comparison Table
-   ## Unknowns & Open Questions
-   ## Recommendation (preliminary — not a decision)
-   ```
-
-8. **Report results:**
-   - Interactive: summarize findings. Do NOT create epics or write code.
-   - Unattended: skip summary, emit status block only:
-     ```
-     PREPLAN_STATUS: { topic, doc_path, unknowns_count, recommendation_confidence: "high|medium|low", source_docs_read: [list] }
-     ```
-
-## Unattended Escalation
-
-When `--unattended` and `recommendation_confidence` is `"low"` (≥5 unresolved unknowns,
-or no clear best-fit approach exists), escalate via the advisor tool rather than stopping.
-Provide the advisor with: topic, unknowns list, comparison table. Resume with advisor's
-direction if given. If advisor cannot resolve, log and skip to next step.
-
-## Rules
-- **Source documents in Step 2 are MANDATORY — research without them is blind**
-- Output is a RESEARCH DOC, not a plan
-- Do not make architectural decisions — surface options with tradeoffs
-- Do not create work items
-- Do not write code
-- Web search is mandatory — don't just analyze the codebase
-- Keep the doc under 200 lines — concise, not exhaustive
-- Unattended: NEVER pause for input; infer and proceed
-- Always list source docs read in the output doc header