@elevasis/sdk 1.22.1 → 1.24.0

package/reference/claude-config/skills/project/SKILL.md

@@ -1,1088 +1,1088 @@
----
-name: project
-description: "Portfolio- and project-level work management -- orientation, intent routing, active projects, milestones, tasks, notes, and resume context -- via the elevasis-sdk project:* CLI."
-argument-hint: "[create | list | work | add | update | status | note | milestone | task | delete] [args]"
-allowed-tools: Bash, Read, Write, Edit, Glob, Grep
----
-
-# Project Management (Portfolio + Per-Project)
-
-`/project` is the primary work-tracking entrypoint for the template. It covers:
-
-- **Portfolio orientation** — bare `/project` invocation shows the landscape of active work and suggests next actions.
-- **Intent routing** — on an opening message that reads as "do work", decide new-vs-resume and route accordingly.
-- **New-work flow** — when the user starts multi-step work without a project, ask: create project, link existing, or standalone.
-- **CRUD surface** — projects, milestones, tasks, notes, checklists via `elevasis-sdk project:*`.
-
-> **Note:** the template no longer ships a separate `/work` skill. Project tasks are the unit of work; use `/project task:*` operations for task lifecycle. Resume context is canonical in `prj_tasks.resume_context` and written via `project:task:save` (never via task-doc frontmatter).
-
-**Usage:**
-
-- `/project` — **Portfolio orientation** (see "Orientation Mode" below). List active/blocked projects, surface most-recently-touched project's `resume_context`, suggest next actions.
-- `/project list` — Same as no-args orientation but just the project table, no suggestions.
-- `/project work <query>` — Fuzzy-match a project/task and print a resume brief (wraps `project:work`).
-- `/project status [<client>]` — Detailed status across all projects, or a single project.
-- `/project create` — Guided project creation (QnA walkthrough).
-- `/project add "Client Name" [--status active] [--value 5000]` — Quick create (flags only, no QnA).
-- `/project update <client> --status on_track` — Update project fields.
-- `/project milestone add <client> "Phase 1" [--due 2026-05-01]` — Add milestone.
-- `/project milestone update <client> "Phase 1" --status completed` — Update milestone.
-- `/project task add <client> "API docs" --milestone "Phase 1" [--type documentation]` — Add task.
-- `/project task update <client> "API docs" --status approved` — Update task.
-- `/project task save <task> --current-state "..."` — Persist `resume_context` (use this, never task-doc frontmatter).
-- `/project note <client> "Kickoff call went well" [--type call_note]` — Add note.
-- `/project notes <client>` — List notes for project.
-- `/project checklist <client> "<milestone>"` — View checklist for a milestone.
-- `/project delete <client>` — Delete project (cascades milestones, tasks, notes).
-
----
-
-## Ambient Vibe Integration
-
-This skill is the landing point for three of the seven vibe intent types. Agents arriving from the ambient layer should behave identically to a direct invocation — vibe is a classifier, not a different code path.
-
-| Vibe intent    | What vibe detected                        | What to do here                                                                                           |
-| -------------- | ----------------------------------------- | --------------------------------------------------------------------------------------------------------- |
-| **Capture**    | "add a task", "remember to", "track this" | Draft the task/note, confirm with user, then `project:task:create` or `project:note:create`               |
-| **Transition** | "done", "stuck", "blocked", "finished"    | Resolve current task from session context, confirm status, then `project:task:update --status <new>`      |
-| **Navigate**   | "focus on", "switch to", "back to"        | Resolve target via `project:resolve <query>` or `project:work <query>`, update scope, narrate new context |
-
-**The full continuity loop** — how work flows across sessions:
-
-```
-vibe (Capture) → /project task:create → task in DB
-                         ↓
-                    work happens
-                         ↓
-              /save → project:task:save → prj_tasks.resume_context
-                         ↓
-              next session: /project work <query>
-                         ↓
-              project:work → resume brief (current state + next steps)
-                         ↓
-              agent picks up exactly where it left off
-```
-
-When a session begins with an ambiguous opening, check `project:work` before asking the user to re-explain. The resume brief is the canonical continuity payload — trust it.
-
----
-
-## Prerequisites
-
-**Run from the project root** (the directory containing `.elevasis`). Before issuing any other commands, run:
-
-```bash
-pnpm elevasis-sdk doctor
-```
-
-If `doctor` fails, **stop immediately and surface the error** — do not retry other commands. Fix the reported issue first (missing `.env`, bad API key, wrong directory, etc.), then re-run `doctor` before proceeding.
-
----
-
-## Invocation Contract
-
-All `elevasis-sdk` commands in this skill use the wrapper script form:
-
-```bash
-pnpm elevasis-sdk <subcommand> [flags]
-```
-
-This form is available from the **project root** — the directory that contains the `.elevasis` marker file. The `.elevasis` file is the project-root anchor: `elevasis-sdk` walks up from its invocation directory until it finds this file, then resolves `.env` and all relative paths from that location. After the project-root refactor, CWD within the project tree matters less — but you must still be somewhere inside the project directory tree.
-
-The long form `pnpm -C operations exec elevasis-sdk <subcommand>` still works and is equivalent. Use the short wrapper form (`pnpm elevasis-sdk`) in all new automation and agent scripts.
-
-**If the `.elevasis` marker is missing**, the CLI exits with:
-
-```
-Not inside an Elevasis project. Run this command from a project directory (an Elevasis project has a .elevasis file at its root).
-```
-
-If you see this error, confirm you are inside a project that was created from the template and has the `.elevasis` marker at its root.
-
----
-
-## Transient Input Files (`tmp/`)
-
-Agents that need to pass structured JSON to `request:submit` or `exec` should write the file to `<projectRoot>/tmp/` rather than to an arbitrary path. The `tmp/` directory is tracked in git (via `tmp/.gitkeep`) but its contents are gitignored, so transient files never pollute the commit history.
-
-**Workflow:**
-
-1. Write the input payload to `tmp/<descriptive-name>.json` from the project root.
-2. Pass the path to the CLI command using the relative form — it resolves against the project root automatically:
-
-   ```bash
-   pnpm elevasis-sdk request:submit -f tmp/request-report.json --cleanup-input
-   pnpm elevasis-sdk exec my-workflow -f tmp/exec-payload.json --cleanup-input
-   ```
-
-3. The `--cleanup-input` flag deletes the file automatically after a successful command. On failure the file is left intact for inspection.
-
-**Safety guard:** `--cleanup-input` only deletes files that are under `<projectRoot>/tmp/`. If the resolved path is outside `tmp/`, the CLI prints a warning to stderr and leaves the file untouched. Files passed via `--input <json>` (inline JSON, no file) are never affected.
-
----
-
-## Orientation Mode (bare `/project` invocation)
-
-When invoked without a subcommand, enter **project mode**: present the portfolio landscape and offer next actions. This is the hub — distinct from `project:work <id>`, which is single-task resume.
-
-### Flow
-
-1. **List active work** — `--status` accepts exactly one value. Run two calls and merge the results:
-
-   ```bash
-   pnpm elevasis-sdk project:list --status active --pretty
-   pnpm elevasis-sdk project:list --status blocked --pretty
-   ```
-
-   Show a compact table: client, status, last-touched.
-
-2. **Surface most-recently-touched project's resume context** — identify the top row (most recent `updated_at`), then print its latest task's `resume_context` (via `project:task:list` → pick most recent in-progress task → `project:task:resume <id> --pretty`). Keep it short: current state + next steps, no files dump.
-
-3. **List available operations** — a one-liner reminder of what the user can do from here:
-
-   ```
-   Available: /project status <client> | /project work <query> | /project task add | /project note | /project create
-   ```
-
-4. **Offer next-action suggestions** — 1-3 concrete follow-ups driven by what's on screen. Examples:
-   - "3 active projects, 1 blocked (Acme — waiting on credentials). Resume Beta (most recent)?"
-   - "No active projects. Create one with `/project create`?"
-   - "Acme's Phase 2 has 2 overdue tasks. Open status?"
-
-### Presentation template
-
-```
-Project Portfolio (2 active, 1 blocked)
-=======================================
-| Client       | Status   | Last touched |
-|--------------|----------|--------------|
-| Beta LLC     | on_track | 2h ago       |
-| Acme Corp    | blocked  | 1d ago       |
-| Gamma Inc    | active   | 4d ago       |
-
-Most recent: Beta LLC — API integration
-  Current: Endpoint wired, tests passing locally.
-  Next:    Deploy to staging, verify webhook delivery.
-
-Next? Resume Beta | Status Acme (why blocked?) | Create new project
-```
-
----
-
-## Intent Detection (opening-message routing)
-
-Apply this heuristic on the user's first substantive message of a session, before branching into any other operation. The goal: distinguish **resume existing work** from **start new work**, and confirm before committing.
-
-### Signals → classification
-
-| Signal in opening message                                                                       | Classify as                        |
-| ----------------------------------------------------------------------------------------------- | ---------------------------------- |
-| References a specific file path, PR, commit, or existing task/milestone/project by name         | **Resume**                         |
-| "Continue", "pick up", "finish", "the thing we were doing", "yesterday's work"                  | **Resume**                         |
-| "Let's build X", "new Y", "I want to add Z", "start a new project", "help me with a new client" | **New work**                       |
-| Single-file cosmetic ask: "fix typo in X", "rename Y", "update dep Z"                           | **Trivial** (skip new-work prompt) |
-| Ambiguous / discussion only                                                                     | Ask before routing                 |
-
-### Routing
-
-- **Resume** — run `project:work <query>` with the extracted entity (file path stem, project name, task keywords). If one clean match, print the brief and ask "Continue here?". If multiple matches, show top 3 and ask which. If zero matches, fall back to orientation mode with a note ("No match for 'acme webhook'; here's the landscape").
-- **New work** — enter the **New-Work Project Decision** flow (next section).
-- **Trivial** — proceed directly; do not prompt about projects. Small single-file edits don't need portfolio tracking.
-- **Ambiguous** — reply with orientation mode output and ask "Resume something, or start new?"
-
-**Always confirm before branching.** A misclassified "new" routed as "resume" burns trust faster than a one-line confirmation.
-
----
-
-## New-Work Project Decision
-
-When intent detection classifies the opening ask as **new multi-step work** and the conversation has no established project context, ask this question before writing any code:
-
-```
-Scope this work under a project?
-  a) Create new project "<proposed-slug>"     (recommended for multi-step work)
-  b) Link to existing "<closest-match>"       (if fuzzy-search finds one)
-  c) Standalone — no project tracking         (one-off / trivial)
-```
-
-### Rules
-
-- **Default to (a)** when the ask is multi-step (multiple files, multiple sessions expected, client-facing deliverable).
-- **Offer (b)** only when `project:list` fuzzy-matches produce a plausible candidate (name similarity, shared entity references). If nothing matches, omit the option — do not invent one.
-- **Skip the prompt entirely** for trivial single-file edits — intent detection's "Trivial" branch handles this.
-- **Propose a slug** by extracting the core noun from the ask. "Build a lead scraper for Acme" → `acme-lead-scraper`. User can override.
-
-### After the decision
-
-- **(a) Create** — invoke the guided `/project create` QnA flow (see below), pre-filling the name/description from the ask.
-- **(b) Link** — run `project:resolve <query>` to confirm the match, then add a task to that project for the new work via `project:task:create`.
-- **(c) Standalone** — proceed without project tracking. Do not create orphan task docs; standalone work leaves no artifact.
-
----
-
-## Connection
-
-### Environment
-
-```bash
-# .env (project root)
-ELEVASIS_PLATFORM_KEY=sk_...
-```
-
-`ELEVASIS_PLATFORM_KEY` is read from the environment automatically. The SDK CLI walks up
-directories to find `.env`, so it resolves correctly from both the project root and
-`operations/`. `ELEVASIS_API_URL` defaults to `https://api.elevasis.io` (production). Set
-`ELEVASIS_API_URL=http://localhost:5170` to target a local API instance, or set
-`NODE_ENV=development` if the SDK CLI respects that convention.
-
-No `--prod` flag is needed — the template targets production by default via `ELEVASIS_API_URL`.
-
-### CLI Invocation
-
-All project operations go through the `elevasis-sdk` CLI. Use the wrapper script form from the
-project root:
-
-```bash
-pnpm elevasis-sdk project:<command> [args]
-```
-
-The long form `pnpm -C operations exec elevasis-sdk project:<command> [args]` is equivalent and
-still works — use it if the wrapper script is not yet available in an older project.
-
-Organization scoping is handled server-side via `ELEVASIS_PLATFORM_KEY`. No org header or
-`--org` flag is required.
-
----
-
-## Database Schema
-
-### Tables
-
-| Table            | Purpose                      | Key Columns                                                           |
-| ---------------- | ---------------------------- | --------------------------------------------------------------------- |
-| `prj_projects`   | Client projects/contracts    | name, kind, status, description, contract_value, start/end dates      |
-| `prj_milestones` | Phases within projects       | name, status, due_date, sequence, completed_at, checklist             |
-| `prj_tasks`      | Work items within milestones | name, status, type, due_date, milestone_id, resume_context, checklist |
-| `prj_notes`      | Meeting notes, updates       | type, content, summary, occurred_at, task_id, milestone_id            |
-
-### Project Kind
-
-| Kind                | Description                      |
-| ------------------- | -------------------------------- |
-| `client_engagement` | External client delivery project |
-| `internal`          | Internal project work            |
-| `research`          | Research and investigation       |
-| `other`             | Catch-all                        |
-
-Default filter for `/project` operations is `kind=client_engagement` unless the user's intent
-is clearly internal or `--kind` is specified explicitly.
-
-### Status Values
-
-| Entity    | Values                                                                                                                   |
-| --------- | ------------------------------------------------------------------------------------------------------------------------ |
-| Project   | `active`, `on_track`, `at_risk`, `blocked`, `completed`, `paused`                                                        |
-| Milestone | `upcoming`, `in_progress`, `completed`, `overdue`, `blocked`                                                             |
-| Task      | `planned`, `in_progress`, `blocked`, `completed`, `cancelled`, `submitted`, `approved`, `rejected`, `revision_requested` |
-| Note type | `call_note`, `status_update`, `issue`, `blocker`, `agent_learning`                                                       |
-| Task type | `documentation`, `code`, `report`, `design`, `refactor`, `feature`, `bug`, `research`, `other`                           |
-
-### Checklist Shape
-
-Both `prj_milestones.checklist` and `prj_tasks.checklist` store a JSONB array of objects with this shape:
-
-```json
-[
-  { "id": "uuid", "label": "Item label", "completed": false },
-  { "id": "uuid", "label": "Another item", "completed": true }
-]
-```
-
-The CLI `--checklist` flag on both `project:milestone:update` and `project:task:update` performs a **full replace** — the entire array is replaced with the JSON you supply. There are no item-level add/toggle/remove flags. To mutate a single item:
-
-1. Read the current checklist (via psql or `project:task:get` / `project:milestone:list`)
-2. Mutate the array in memory (append, flip `completed`, filter out)
-3. Write the entire array back via `--checklist '<json>'`
-
-To clear a checklist: `--checklist '[]'`.
-
-### Foreign Keys
-
-- `prj_milestones.project_id` → `prj_projects.id` (CASCADE)
-- `prj_tasks.project_id` → `prj_projects.id` (CASCADE)
-- `prj_tasks.milestone_id` → `prj_milestones.id` (SET NULL)
-- `prj_tasks.parent_task_id` → `prj_tasks.id` (SET NULL) — for subtasks
-- `prj_notes.project_id` → `prj_projects.id` (CASCADE)
-- `prj_notes.task_id` → `prj_tasks.id` (SET NULL)
-- `prj_notes.milestone_id` → `prj_milestones.id` (SET NULL)
-- `prj_projects.deal_id` → `acq_deals.id` (SET NULL)
-- `prj_projects.client_id` → `clients.id` (SET NULL) — clients-hub link; set via `--client`
-- `prj_projects.client_company_id` → `acq_companies.id` (SET NULL) — legacy direct-company link; set via `--client-company-id`
-
----
-
-## Operations
-
-### `list` — Portfolio Table (no orientation extras)
-
-For a plain project table without resume context or next-action suggestions (use the bare `/project` orientation mode above when the user wants the full hub experience):
-
-```bash
-pnpm elevasis-sdk project:list --kind client_engagement --pretty
-
-# Filter by client (accepts the client's name or UUID)
-pnpm elevasis-sdk project:list --client "Acme" --pretty
-```
-
-Present as:
-
-```
-Project Portfolio
-=================
-| Client             | Status   | Kind               | Value   |
-|--------------------|----------|--------------------|---------|
-| Acme Corp          | on_track | client_engagement  | $5,000  |
-| Beta LLC           | active   | client_engagement  | $3,500  |
-```
-
-For progress detail (milestones/tasks), follow up with `project:milestone:list` and
-`project:task:list` per project.
-
-### `work <query>` — Resume Brief
-
-Wraps `project:work` for fuzzy-matched resume. Used by intent-detection "resume" classification.
-
-```bash
-pnpm elevasis-sdk project:work <query>
-```
-
-Prints the matched project/task brief with current `resume_context`. If multiple matches, show top 3 and ask which.
-
-### `status [<client>]` — Detailed Status
-
-If client specified, show full detail for that project. If not, show all.
-
-**Resolve the project first** (see Client Inference below), then:
-
-```bash
-# Get project details
-pnpm elevasis-sdk project:get <project-id>
-
-# Milestones (ordered by sequence)
-pnpm elevasis-sdk project:milestone:list --project <project-id> --pretty
-
-# Tasks (grouped by milestone)
-pnpm elevasis-sdk project:task:list --project <project-id> --pretty
-
-# Recent notes
-pnpm elevasis-sdk project:note:list --project <project-id> --pretty
-```
-
-Present as:
-
-```
-Acme Corp — On Track
-=====================
-Description: Automation project for AI-powered workflows
-Contract: $5,000 | Started: 2026-04-01 | Target: 2026-06-01
-
-Milestones (2/4 complete)
-  [x] Phase 1: Discovery (completed)
-  [>] Phase 2: Implementation (in_progress, due 2026-05-01)
-  [ ] Phase 3: Testing (upcoming, due 2026-05-15)
-  [ ] Phase 4: Launch (upcoming, due 2026-06-01)
-
-Tasks (3/8 approved)
-  Phase 2: Implementation
-    [approved] Requirements doc (documentation)
-    [in_progress] API integration (code, due 2026-05-01)
-    [planned] User guide (documentation)
-
-Recent Notes
-  2026-04-05 [call_note] Weekly check-in — on track, API work started
-  2026-04-01 [call_note] Kickoff — confirmed scope and timeline
-```
-
-#### Tenant Skill Bindings Footer
-
-After the normal status output, append a short footer when the project kind, tags, name,
-description, milestones, or active tasks map to a known tenant domain such as `prospecting`,
-`crm`, `outreach`, `finance`, or `support`.
-
-Resolve the best domain from explicit metadata first (`tags`, `kind`, domain fields), then from
-project/task text. If no clear domain maps, omit the footer.
-
-Footer shape:
-
-```text
-Related skill bindings
-- Domain: <domain>
-- Read or change business profile: /knowledge <domain>
-- Layering primer: node_modules/@elevasis/sdk/reference/spine/spine-primer.md
-```
-
-External projects do not expose monorepo-only architecture commands. Never suggest `/org-os`,
-`/om-spine`, or monorepo paths in this footer.
-
-### `create` — Guided Project Creation (QnA Flow)
-
-Interactive walkthrough that collects project details step by step using `AskUserQuestion`,
-then creates the project, milestones, and kickoff note in one go.
-
-**Step 1: Project basics**
-
-Ask via `AskUserQuestion` (single question, free-text via "Other"):
-
-```
-What's the project name and a brief description of the scope?
-```
-
-The user will reply with something like "Acme Corp automation -- building a lead gen pipeline
-with email outreach". Parse the name and description from the response.
-
-**Step 2: Deal/company linking**
-
-Search for matching records using read-only psql (the psql role is SELECT-only).
-
-Note: `SUPABASE_READONLY_URL` must be set in the template `.env` for psql queries. If not
-set, skip this step and advise the user to link via the Command Center UI after creation.
-
-If `SUPABASE_READONLY_URL` is available, query using `<ORG_ID>` from the platform key's
-associated organization (you can retrieve it via `project:list` output which includes
-`organization_id`):
-
-```bash
-source .env && psql "$SUPABASE_READONLY_URL" -c \
-  "SELECT id, name, domain FROM acq_companies WHERE organization_id = '<ORG_ID>' AND name ILIKE '%<parsed_name>%' ORDER BY updated_at DESC LIMIT 5;"
-
-source .env && psql "$SUPABASE_READONLY_URL" -c \
-  "SELECT d.id, d.contact_email, d.cached_stage, c.name AS company_name
-   FROM acq_deals d
-   LEFT JOIN acq_contacts ct ON ct.id = d.contact_id
-   LEFT JOIN acq_companies c ON c.id = ct.company_id
-   WHERE d.organization_id = '<ORG_ID>'
-     AND (d.contact_email ILIKE '%<parsed_name>%' OR c.name ILIKE '%<parsed_name>%')
-   ORDER BY d.updated_at DESC LIMIT 5;"
-```
-
-If matches found, ask via `AskUserQuestion`:
-
-```
-Found matching records. Link this project?
-- [Company: Acme Corp (acme.com)]
-- [Deal: john@acme.com (closed_won)]
-- No linking
-```
-
-If no matches, skip silently.
-
-**Step 3: Contract details**
-
-Ask via `AskUserQuestion`:
-
-```
-Contract details?
-- Options with previews showing common structures:
-  - "Fixed project" -- one-time fee
-  - "Monthly retainer" -- recurring
-  - "Hourly" -- time & materials
-  - Other
-```
-
-Then follow up for the value, start date, and target end date. These can be collected in a
-single question:
-
-```
-What's the contract value, start date, and target end date?
-(e.g., "$5000, starting today, targeting June 1")
-```
-
-Parse natural language dates ("today", "next Monday", "June 1") into ISO format.
-
-**Step 4: Milestone suggestions**
-
-Based on the project description from Step 1, suggest 3-5 milestones that fit the scope. Use
-your understanding of the project to propose relevant phases.
-
-**Guidelines for milestone suggestions:**
-
-- For automation/integration projects: Discovery, Implementation, Testing, Launch
-- For content/marketing projects: Strategy, Content Creation, Review, Distribution
-- For consulting/advisory: Assessment, Recommendations, Implementation Support
-- Always include a Discovery/Kickoff phase first and a Launch/Handoff phase last
-- Suggest due dates spaced evenly between start and target end date
-
-Present via `AskUserQuestion` with previews:
-
-```
-Suggested milestones for "<project name>". Select which to include:
-(multiSelect: true)
-
-- Discovery (due <start + 2 weeks>)
-- Implementation (due <midpoint>)
-- Testing (due <end - 2 weeks>)
-- Launch (due <end date>)
-```
-
-The user can select which ones to keep, modify via "Other", or skip entirely.
-
-**Step 5: Kickoff note**
-
-Ask via `AskUserQuestion`:
-
-```
-Add a kickoff note? This captures initial context (scope confirmed, key contacts, etc.)
-- Yes, let me type one
-- Skip for now
-```
-
-If yes, collect the note content.
-
-**Step 6: Execute**
-
-Create everything in sequence using the CLI:
-
-```bash
-# 1. Create project
-# If the user mentioned a client by name, resolve it first:
-#   pnpm elevasis-sdk client:resolve "<client name>"
-# Then pass the name (or UUID) via --client
-pnpm elevasis-sdk project:create \
-  --name "<name>" \
-  --kind client_engagement \
-  --status active \
-  --description "<desc>" \
-  --client "<client-name-or-uuid>"
-
-# 2. Create milestones
-pnpm elevasis-sdk project:milestone:create \
-  --project <project-id> \
-  --name "<milestone-name>" \
-  --status upcoming \
-  --due-date <date>
-
-# 3. Create kickoff note (if provided)
-pnpm elevasis-sdk project:note:create \
-  --project <project-id> \
-  --content "<note-content>" \
-  --type call_note
-```
-
-If `--company` or `--deal` linking was confirmed, use `project:update` after creation:
-
-```bash
-pnpm elevasis-sdk project:update <project-id> --client "<client-name-or-uuid>"
-```
-
-To remove a client link later:
-
-```bash
-pnpm elevasis-sdk project:update <project-id> --clear-client
-```
-
-Do NOT pass `--client ""` to clear — an empty string is rejected as ambiguous. Use `--clear-client` instead.
-
-**Step 7: Summary**
-
-Show the created project:
-
-```
-Project Created
-===============
-Name: Acme Corp Automation
-Kind: client_engagement
-Status: active
-Timeline: 2026-04-07 → 2026-06-01
-
-Milestones:
-  1. Discovery (upcoming, due 2026-04-21)
-  2. Implementation (upcoming, due 2026-05-08)
-  3. Testing (upcoming, due 2026-05-22)
-  4. Launch (upcoming, due 2026-06-01)
-
-Kickoff Note: Confirmed scope -- lead gen pipeline with email outreach.
-
-Next: /project status acme
-```
-
----
-
-### `add "<name>" [options]` — Quick Create Project (No QnA)
-
-**Options:**
-
-- `--status <status>` (default: `active`)
-- `--value <number>` — contract value (informational, stored in description or notes)
-- `--start <date>` — start date
-- `--end <date>` — target end date
-- `--description "<text>"` — scope description
-- `--kind <kind>` (default: `client_engagement`)
-
-```bash
-pnpm elevasis-sdk project:create \
-  --name "<name>" \
-  --kind client_engagement \
-  --status active \
-  --description "<desc>"
-```
-
-**After creating:** Show the created project and suggest next steps (add milestones).
-
-### `update <client> [options]` — Update Project
-
-**Options:** `--status`, `--name`, `--description`, `--client <id-or-name>`, `--clear-client`, `--client-company-id <uuid>`
-
-First resolve the client (see Client Inference below), then:
-
-```bash
-pnpm elevasis-sdk project:update <project-id> --status <status>
-```
-
-### `milestone add <client> "<name>" [options]` — Add Milestone
-
-**Options:**
-
-- `--status <status>` (default: `upcoming`)
-- `--due <date>` — due date (ISO)
-- `--description "<text>"`
-
-```bash
-pnpm elevasis-sdk project:milestone:create \
-  --project <project-id> \
-  --name "<name>" \
-  --status upcoming \
-  --due-date <date>
-```
-
-### `milestone update <client> "<milestone>" [options]` — Update Milestone
-
-Resolve milestone by name (see Client Inference for project resolution, then use
-`project:milestone:list` to find the milestone ID by partial name match):
-
-```bash
-# List milestones to find ID
-pnpm elevasis-sdk project:milestone:list --project <project-id>
-
-# Update status
-pnpm elevasis-sdk project:milestone:update <milestone-id> --status completed
-
-# Update checklist (full replace)
-pnpm elevasis-sdk project:milestone:update <milestone-id> \
-  --checklist '[{"id":"uuid","label":"Item label","completed":false}]'
-```
-
-If status changes to `completed`, the API auto-sets `completed_at`.
-
-**Options:** `--status`, `--due-date`, `--name`, `--checklist <json>`
-
-**Checklist mutations:** The CLI performs a full replace. To add, toggle, or remove an item:
-
-1. Read the current checklist via psql or `project:milestone:list`
-2. Mutate the array in memory
-3. Write back the full array via `--checklist '<json>'`
-
-### `checklist <client> "<milestone>"` — View Checklist
-
-Resolve the milestone, then read its `checklist` JSONB column via read-only psql.
-
-Note: Requires `SUPABASE_READONLY_URL` in the template `.env`. If not set, advise the user
-to view the checklist via the Command Center UI, or add `SUPABASE_READONLY_URL` to `.env`.
-
-```bash
-source .env && psql "$SUPABASE_READONLY_URL" -c \
-  "SELECT m.name, m.checklist
-   FROM prj_milestones m
-   WHERE m.project_id = '<project_id>' AND m.name ILIKE '%<name>%'
-   LIMIT 1;"
-```
-
-The `checklist` column is a JSONB array of objects:
-
-```json
-[
-  { "id": "uuid", "label": "Set up dev environment", "completed": true },
-  { "id": "uuid", "label": "Draft scope document", "completed": false }
-]
-```
-
-Present as:
-
-```
-Onboarding & Scope — Checklist (2/3 complete)
-  [x] Set up dev environment
-  [x] Schedule kickoff call
-  [ ] Draft scope document
-```
-
-### `task add <client> "<name>" [options]` — Add Task
-
-**Options:**
-
-- `--milestone "<name>"` — link to a milestone (resolve by name first)
-- `--status <status>` (default: `planned`)
-- `--type <type>` (default: `other`)
-- `--due <date>`
-- `--description "<text>"`
-- `--checklist <json>` — initial checklist items
-
-```bash
-# Resolve milestone ID first if --milestone provided
-pnpm elevasis-sdk project:milestone:list --project <project-id>
-
-# Create task
-pnpm elevasis-sdk project:task:create \
-  --project <project-id> \
-  --title "<name>" \
-  --status planned \
-  --type <type> \
-  --milestone <milestone-id>
-
-# Create task with initial checklist
-pnpm elevasis-sdk project:task:create \
-  --project <project-id> \
-  --title "<name>" \
-  --checklist '[{"id":"1","label":"Step one","completed":false}]'
-```
-
-### `task update <client> "<task>" [options]` — Update Task
-
-Resolve task by name within the project using `project:task:list`, then update by positional ID:
-
-```bash
-# Find the task ID
-pnpm elevasis-sdk project:task:list --project <project-id>
-
-# Update status (positional <id> comes first, then flags)
-pnpm elevasis-sdk project:task:update <task-id> --status <status>
-
-# Update checklist (full replace)
-pnpm elevasis-sdk project:task:update <task-id> \
-  --checklist '[{"id":"uuid","label":"Step","completed":false}]'
-
-# Clear checklist
-pnpm elevasis-sdk project:task:update <task-id> --checklist '[]'
-```
-
-**Options:** `--status`, `--title`, `--milestone`, `--description`, `--checklist <json>`
-
-**Important:** The task ID is a **positional argument** — it comes immediately after `project:task:update`, before any flags. There is no `--task-id` flag.
-
-If status changes to `approved`, the API auto-sets `completed_at`.
-
-**Checklist mutations:** Same full-replace semantics as milestones. To add, toggle, or remove an item:
-
-1. Read the current task via `project:task:get <task-id>`
-2. Mutate the checklist array in memory
-3. Write back the full array via `project:task:update <task-id> --checklist '<json>'`
-
-### `note <client> "<content>" [options]` — Add Note
-
-**Options:**
-
-- `--type <type>` (default: `call_note`) — see Note Type Selection below
-- `--task <task-id>` — attach to a specific task (UUID)
-- `--milestone <milestone-id>` — attach to a milestone (UUID)
-
-```bash
-pnpm elevasis-sdk project:note:create \
-  --project <project-id> \
-  --content "<content>" \
-  --type <type>
-
-# Attach to a task
-pnpm elevasis-sdk project:note:create \
-  --project <project-id> \
-  --content "<content>" \
-  --type agent_learning \
-  --task <task-id>
-```
-
-### `notes <client>` — List Notes
-
-```bash
-pnpm elevasis-sdk project:note:list --project <project-id> --pretty
-```
-
-### `delete <client>` — Delete Project
-
-**ALWAYS confirm before deleting.** Show what will be deleted (milestone count, task count,
-note count) using the list commands, then ask for confirmation.
-
-```bash
-pnpm elevasis-sdk project:delete <project-id>
-```
-
-Cascade deletes all milestones, tasks, and notes.
-
----
-
-## Notes Management
-
-Notes (`prj_notes`) are the durable record of project-scoped signals — conversation outcomes, blockers, status updates, and agent-captured learnings. They complement `resume_context` (which is task-scoped, temporal, and agent-facing) by providing a human-readable audit trail.
-
-### When to Create a Note vs. Fall Through to Global Auto-Memory
-
-**Create a project note** when all of the following hold:
-
-- A specific project and (optionally) task is in scope in the current session
-- The content is project-bound — it would only matter to someone working on this project
-- The content is durable — it should be visible in future status views, not just the current session
-
-**Fall through to global auto-memory** when:
-
-- No active project or task can be resolved from session context (see Project-Scope Resolution below)
-- The content is cross-project or platform-wide (e.g., a general TypeScript pattern, a reusable workflow insight)
-- The signal is ephemeral and not worth a DB write (e.g., "reminder: run linter before committing")
-
-When in doubt and no project context is resolvable, global auto-memory is the safe default.
-
-### Note Type Selection Guide
-
-| Type             | Use when                                                                                                                                                                                                                                                   |
-| ---------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `call_note`      | Transcribing or summarizing a live client / stakeholder call. Include key decisions and action items.                                                                                                                                                      |
-| `status_update`  | Milestone-level progress worth flagging to a human operator: phase complete, substantial deliverable, direction change. Don't emit one every session turn.                                                                                                 |
-| `issue`          | Bug, regression, unexpected failure, contract mismatch. Include reproduction context.                                                                                                                                                                      |
-| `blocker`        | Work cannot proceed without external action: decision needed, credential missing, upstream fix required.                                                                                                                                                   |
-| `agent_learning` | Project-bound knowledge the agent discovered: API quirks, client conventions, deploy gotchas, rate limits. Captures "things a future agent resuming this project should know" so it travels with the project instead of living only in global auto-memory. |
-
-### Project-Scope Resolution
-
-Before writing a note, the agent resolves which project (and optionally task) to attach it to:
-
-1. **Explicit in session** — a project/task UUID was mentioned or returned by a recent `project:work <id>` or `project:task:*` invocation in this session. Use it directly.
-2. **Implicit from task list** — if step 1 yields nothing, run (`--status` accepts one value; run both calls and merge):
-
-   ```bash
-   pnpm elevasis-sdk project:list --status active --pretty
-   pnpm elevasis-sdk project:list --status blocked --pretty
-   pnpm elevasis-sdk project:task:list --project <top-project-id> --status in_progress
-   ```
-
-   Use the most-recently-touched project's in-progress task if there is exactly one. If multiple in-progress tasks exist, prefer the one that most closely matches the current conversation topic.
-
-3. **Prompt the user** — if steps 1 and 2 are both ambiguous or empty, ask once: "Which project should I attach this note to? (project UUID or slug, or 'skip' to use global auto-memory)."
-
-4. **Fall through** — if the user says "skip" or no resolution is possible, write to global auto-memory and continue.
-
-Resolution is **agent-side at call time** — there is no middleware or automatic routing layer. The agent infers context from the session.
-
-### Agent Workflow: Capturing an `agent_learning`
-
-When the agent recognizes a project-scoped learning (an API quirk, an undocumented constraint, a client convention discovered during implementation):
-
-1. Recognize the signal — e.g., "Apify actor X rate-limits at 20 rps", "client requires ISO dates, not timestamps", "staging env needs VPN before webhook delivery works"
-2. Resolve active project and task from session context (see above)
-3. Write the note:
-
-   ```bash
-   pnpm elevasis-sdk project:note:create \
-     --project <project-id> \
-     --type agent_learning \
-     --task <task-id> \
-     --content "<concise, factual statement of what was learned>"
-   ```
-
-   Omit `--task` if no specific task is in scope (e.g., the learning is milestone-level or project-level).
-
-4. If no project can be resolved, write the learning to global auto-memory instead.
-
-**Content style for `agent_learning` notes:** write as a factual assertion another agent could act on — "Apify actor X rate-limits at 20 rps; add a 50ms delay between calls" rather than "I noticed that Apify seems slow." Keep it under 200 characters when possible.
-
----
-
-## Status Transitions from Other Skills
-
-The `/project` skill is the canonical reference for how lifecycle signals from other skills map to `project:task:update` status transitions. The behavioral hooks live in the emitting skills; this section documents the contract.
-
-### Transition Map
-
-| Signal                                   | Source skill   | Transition                             | CLI call                                           |
-| ---------------------------------------- | -------------- | -------------------------------------- | -------------------------------------------------- |
-| Successful `git push` via `/deploy`      | `deploy/`      | linked task → `submitted`              | `project:task:update <task-id> --status submitted` |
-| `/save` catches "I'm stuck" / blocker    | `save/`        | linked task → `blocked` + blocker note | `project:task:update <task-id> --status blocked`   |
-| User/agent says "I'm done" / "task done" | `/project` NLM | active task → `completed`              | `project:task:update <task-id> --status completed` |
-
-### "I'm Done" Recognition (Natural Language Mode)
-
-When the user or agent says any of the following in the context of a project task, treat it as a `completed` transition signal:
-
-- "I'm done", "I'm finished", "done with this", "that's done"
-- "task complete", "task finished", "mark as complete"
-- "wrap this up", "close this task", "finalize this"
-
-Resolve the active task from session context (see Project-Scope Resolution), confirm once ("Mark `<task name>` as completed?"), then fire:
-
-```bash
-pnpm elevasis-sdk project:task:update <task-id> --status completed
-```
-
-If no task resolves, ask the user which task to complete rather than silently skipping.
-
-### Transition Rules (All Skills)
-
-These rules apply to every skill that emits a transition:
-
-- **Explicit task ID only** — use the UUID resolved from session context. Never guess.
-- **Best-effort** — if the CLI returns non-2xx, emit a warning line and continue. Do not block the primary operation.
-- **Idempotent** — if the task is already in the target status, the update is a no-op. Fire it anyway.
-- **Confirm before `completed`** — always ask once before marking a task done, since this is hard to undo semantically.
-
-For full implementation details, see the emitting skills: `deploy/SKILL.md` (Step 8) and `save/SKILL.md` (Steps 5, 5a).
-
----
-
-## Client Inference
-
-Resolve which project the user means using these rules (in priority order):
-
-1. **Exact ID** — if the argument looks like a UUID, use it directly with `project:get`
-2. **Name match** — list projects filtered by `client_engagement` kind, then match by name:
-   ```bash
-   pnpm elevasis-sdk project:list --kind client_engagement
-   ```
-   Filter results by partial name match against the user's input (case-insensitive).
-3. **Single active project** — if only one non-completed `client_engagement` project exists,
-   use it without asking
-4. **Context from conversation** — if the conversation has been discussing a specific client,
-   use that one
-5. **Ambiguous** — if multiple match or none match, list projects and ask
-
----
-
-## Natural Language Mode
-
-When args don't match any command pattern, infer intent from natural language:
-
-| User says                                                      | Inferred operation                                                                                                                                    |
-| -------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- |
-| "how's acme doing"                                             | `status acme`                                                                                                                                         |
-| "add a note: weekly call, everything on track"                 | `note <active-project> "weekly call, everything on track"`                                                                                            |
-| "mark phase 2 complete for acme"                               | `milestone update acme "phase 2" --status completed`                                                                                                  |
-| "approve the API docs task"                                    | `task update <project> "API docs" --status approved`                                                                                                  |
-| "new client: Beta LLC, $3500/mo, starting may 1"               | `add "Beta LLC" --value 3500 --start 2026-05-01`                                                                                                      |
-| "what's overdue?"                                              | List milestones where `due_date < now() AND status != 'completed'`                                                                                    |
-| "block acme, waiting on client credentials"                    | `update acme --status blocked`                                                                                                                        |
-| "show checklist for phase 2"                                   | Read and display checklist for milestone "phase 2" via psql or `project:milestone:list`                                                               |
-| "add 'deploy staging' to task X's checklist"                   | Read task X's checklist → append `{id: uuid, label: "deploy staging", completed: false}` → `project:task:update <id> --checklist '<json>'`            |
-| "mark 'deploy staging' done on task X"                         | Read task X's checklist → flip `completed` on matching item → `project:task:update <id> --checklist '<json>'`                                         |
-| "clear the checklist on task X"                                | `project:task:update <task-id> --checklist '[]'`                                                                                                      |
-| "add checklist item to onboarding milestone: review scope doc" | Read milestone checklist → append item → `project:milestone:update <id> --checklist '<json>'`                                                         |
-| "I'm done" / "task complete" / "done with this"                | Resolve active task → confirm → `project:task:update <task-id> --status completed`                                                                    |
-| "save: Apify actor X rate-limits at 20rps"                     | Resolve active project/task → `project:note:create --project <id> --type agent_learning --task <task-id> "Apify actor X rate-limits at 20rps"`        |
-| "remember: client requires ISO dates, not timestamps"          | Resolve active project/task → `project:note:create --project <id> --type agent_learning --task <task-id> "client requires ISO dates, not timestamps"` |
-
-**Checklist note:** all checklist mutations use the read-modify-write pattern. The CLI has no item-level flags (`--add-item`, `--toggle`, `--remove-item` do not exist). Always read the current state, mutate the array, then write the full array back.
-
----
-
-## Bulk Operations
-
-### Add Multiple Milestones
-
-For creating a standard milestone set, run `project:milestone:create` in sequence:
-
-```bash
-for name in "Discovery" "Implementation" "Testing" "Launch"; do
-  pnpm elevasis-sdk project:milestone:create \
-    --project <project-id> \
-    --name "$name" \
-    --status upcoming
-done
-```
-
-### Overdue Report
-
-Query via read-only psql for milestones past their due date. Requires `SUPABASE_READONLY_URL`
-in the template `.env`. Replace `<ORG_ID>` with the organization ID associated with your
-`ELEVASIS_PLATFORM_KEY` (visible in `project:list` output as `organization_id`). If
-`SUPABASE_READONLY_URL` is not configured, use `project:milestone:list` per project and filter
-manually.
-
-```bash
-source .env && psql "$SUPABASE_READONLY_URL" -c \
-  "SELECT
-     p.name AS project,
-     m.name AS milestone,
-     m.due_date,
-     m.status,
-     (CURRENT_DATE - m.due_date::date) AS days_overdue
-   FROM prj_milestones m
-   JOIN prj_projects p ON p.id = m.project_id
-   WHERE p.organization_id = '<ORG_ID>'
-     AND m.due_date < CURRENT_DATE
-     AND m.status NOT IN ('completed')
-   ORDER BY m.due_date;"
-```
-
----
-
-## Client Linking
-
-Projects can be linked to a client in the clients hub. When a user mentions a client by name, resolve it to an ID and use `--client` to wire the project.
-
-### Two linking flags, two different columns
-
-- `--client <name-or-uuid>` — links to a clients-hub record (`prj_projects.client_id`). Use this for new client linkage. Accepts a name (the CLI fuzzy-resolves it) or a UUID.
-- `--client-company-id <uuid>` — legacy direct-company linkage (`prj_projects.client_company_id`); kept for back-compat. UUID only.
-
-These are independent. Do not use `--client-company-id` when you mean `--client`.
-
-### When the user mentions a client by name
-
-1. Resolve the client name to a UUID first (optional but gives an explicit confirmation step):
-   ```bash
-   pnpm elevasis-sdk client:resolve "Acme"
-   ```
-2. Pass the name or UUID to `--client` — the CLI fuzzy-resolves names automatically:
-   ```bash
-   pnpm elevasis-sdk project:create --name "Acme Automation" --kind client_engagement --client "Acme"
-   pnpm elevasis-sdk project:update <project-id> --client "Acme"
-   ```
-3. If the client name matches multiple records, the CLI returns an error listing candidates. In that case, run `client:resolve "Acme"` to disambiguate, then pass the UUID directly.
-
-### Filtering projects by client
-
-```bash
-pnpm elevasis-sdk project:list --client "Acme" --pretty
-```
-
-### Removing a client link
-
-```bash
-pnpm elevasis-sdk project:update <project-id> --clear-client
-```
-
-`--client` and `--clear-client` are mutually exclusive on the same command call.
-
-### Soft recommendation for client engagement projects
-
-When the user creates a `client_engagement` project without mentioning a client, gently note that linking a client is recommended so the project appears in client lineage views. Do not block creation — the flag is optional.
-
-### Client command reference
-
-The full `client:*` surface (list, get, status, resolve) is available via `elevasis-sdk client:*`. The `client:resolve` command mirrors `project:resolve` in shape and is the canonical tool for name-to-ID translation.
-
----
-
-## Safety Rules
-
-1. **Always confirm deletes** — show what will be cascade-deleted before executing
-2. **Organization scoping** — all API calls are automatically org-scoped via
-   `ELEVASIS_PLATFORM_KEY`; every direct psql query MUST include an `organization_id` filter
-   (use `<ORG_ID>` as a placeholder and confirm the value with the user if unknown)
-3. **Read before write** — when updating, show current state before applying changes
-4. **Validate status values** — reject invalid status strings before sending to the CLI
-5. **Kind default** — default to `--kind client_engagement` when user intent is client
-   delivery; use `--kind internal` for internal project work
-6. **Template project assumption** — the template is a single-organization project. All
-   `/project` operations operate within the organization scoped by `ELEVASIS_PLATFORM_KEY`.
-   There is no cross-org capability.
-7. **Checklist full-replace** — the `--checklist` flag replaces the entire array. Always read
-   the current checklist before writing to avoid losing existing items.
-8. **Confirm task completion** — always ask once before marking a task `completed` via "I'm done"
-   recognition. This status is semantically significant and warrants a confirmation step.
-
----
-
-**Last Updated:** 2026-05-08
+---
+name: project
+description: "Portfolio- and project-level work management -- orientation, intent routing, active projects, milestones, tasks, notes, and resume context -- via the elevasis-sdk project:* CLI."
+argument-hint: "[create | list | work | add | update | status | note | milestone | task | delete] [args]"
+allowed-tools: Bash, Read, Write, Edit, Glob, Grep
+---
+
+# Project Management (Portfolio + Per-Project)
+
+`/project` is the primary work-tracking entrypoint for the template. It covers:
+
+- **Portfolio orientation** — bare `/project` invocation shows the landscape of active work and suggests next actions.
+- **Intent routing** — on an opening message that reads as "do work", decide new-vs-resume and route accordingly.
+- **New-work flow** — when the user starts multi-step work without a project, ask: create project, link existing, or standalone.
+- **CRUD surface** — projects, milestones, tasks, notes, checklists via `elevasis-sdk project:*`.
+
+> **Note:** the template no longer ships a separate `/work` skill. Project tasks are the unit of work; use `/project task:*` operations for task lifecycle. Resume context is canonical in `prj_tasks.resume_context` and written via `project:task:save` (never via task-doc frontmatter).
+
+**Usage:**
+
+- `/project` — **Portfolio orientation** (see "Orientation Mode" below). List active/blocked projects, surface most-recently-touched project's `resume_context`, suggest next actions.
+- `/project list` — Same as no-args orientation but just the project table, no suggestions.
+- `/project work <query>` — Fuzzy-match a project/task and print a resume brief (wraps `project:work`).
+- `/project status [<client>]` — Detailed status across all projects, or a single project.
+- `/project create` — Guided project creation (QnA walkthrough).
+- `/project add "Client Name" [--status active] [--value 5000]` — Quick create (flags only, no QnA).
+- `/project update <client> --status on_track` — Update project fields.
+- `/project milestone add <client> "Phase 1" [--due 2026-05-01]` — Add milestone.
+- `/project milestone update <client> "Phase 1" --status completed` — Update milestone.
+- `/project task add <client> "API docs" --milestone "Phase 1" [--type documentation]` — Add task.
+- `/project task update <client> "API docs" --status approved` — Update task.
+- `/project task save <task> --current-state "..."` — Persist `resume_context` (use this, never task-doc frontmatter).
+- `/project note <client> "Kickoff call went well" [--type call_note]` — Add note.
+- `/project notes <client>` — List notes for project.
+- `/project checklist <client> "<milestone>"` — View checklist for a milestone.
+- `/project delete <client>` — Delete project (cascades milestones, tasks, notes).
+
+---
+
+## Ambient Vibe Integration
+
+This skill is the landing point for three of the seven vibe intent types. Agents arriving from the ambient layer should behave identically to a direct invocation — vibe is a classifier, not a different code path.
+
+| Vibe intent    | What vibe detected                        | What to do here                                                                                           |
+| -------------- | ----------------------------------------- | --------------------------------------------------------------------------------------------------------- |
+| **Capture**    | "add a task", "remember to", "track this" | Draft the task/note, confirm with user, then `project:task:create` or `project:note:create`               |
+| **Transition** | "done", "stuck", "blocked", "finished"    | Resolve current task from session context, confirm status, then `project:task:update --status <new>`      |
+| **Navigate**   | "focus on", "switch to", "back to"        | Resolve target via `project:resolve <query>` or `project:work <query>`, update scope, narrate new context |
+
+**The full continuity loop** — how work flows across sessions:
+
+```
+vibe (Capture) → /project task:create → task in DB
+                         ↓
+                    work happens
+                         ↓
+              /save → project:task:save → prj_tasks.resume_context
+                         ↓
+              next session: /project work <query>
+                         ↓
+              project:work → resume brief (current state + next steps)
+                         ↓
+              agent picks up exactly where it left off
+```
+
+When a session begins with an ambiguous opening, check `project:work` before asking the user to re-explain. The resume brief is the canonical continuity payload — trust it.
+
+---
+
+## Prerequisites
+
+**Run from the project root** (the directory containing `.elevasis`). Before issuing any other commands, run:
+
+```bash
+pnpm elevasis-sdk doctor
+```
+
+If `doctor` fails, **stop immediately and surface the error** — do not retry other commands. Fix the reported issue first (missing `.env`, bad API key, wrong directory, etc.), then re-run `doctor` before proceeding.
+
+---
+
+## Invocation Contract
+
+All `elevasis-sdk` commands in this skill use the wrapper script form:
+
+```bash
+pnpm elevasis-sdk <subcommand> [flags]
+```
+
+This form is available from the **project root** — the directory that contains the `.elevasis` marker file. The `.elevasis` file is the project-root anchor: `elevasis-sdk` walks up from its invocation directory until it finds this file, then resolves `.env` and all relative paths from that location. After the project-root refactor, CWD within the project tree matters less — but you must still be somewhere inside the project directory tree.
+
+The long form `pnpm -C operations exec elevasis-sdk <subcommand>` still works and is equivalent. Use the short wrapper form (`pnpm elevasis-sdk`) in all new automation and agent scripts.
+
+**If the `.elevasis` marker is missing**, the CLI exits with:
+
+```
+Not inside an Elevasis project. Run this command from a project directory (an Elevasis project has a .elevasis file at its root).
+```
+
+If you see this error, confirm you are inside a project that was created from the template and has the `.elevasis` marker at its root.
+
+---
+
+## Transient Input Files (`tmp/`)
+
+Agents that need to pass structured JSON to `request:submit` or `exec` should write the file to `<projectRoot>/tmp/` rather than to an arbitrary path. The `tmp/` directory is tracked in git (via `tmp/.gitkeep`) but its contents are gitignored, so transient files never pollute the commit history.
+
+**Workflow:**
+
+1. Write the input payload to `tmp/<descriptive-name>.json` from the project root.
+2. Pass the path to the CLI command using the relative form — it resolves against the project root automatically:
+
+   ```bash
+   pnpm elevasis-sdk request:submit -f tmp/request-report.json --cleanup-input
+   pnpm elevasis-sdk exec my-workflow -f tmp/exec-payload.json --cleanup-input
+   ```
+
+3. The `--cleanup-input` flag deletes the file automatically after a successful command. On failure the file is left intact for inspection.
+
+**Safety guard:** `--cleanup-input` only deletes files that are under `<projectRoot>/tmp/`. If the resolved path is outside `tmp/`, the CLI prints a warning to stderr and leaves the file untouched. Files passed via `--input <json>` (inline JSON, no file) are never affected.
+
+---
+
+## Orientation Mode (bare `/project` invocation)
+
+When invoked without a subcommand, enter **project mode**: present the portfolio landscape and offer next actions. This is the hub — distinct from `project:work <id>`, which is single-task resume.
+
+### Flow
+
+1. **List active work** — `--status` accepts exactly one value. Run two calls and merge the results:
+
+   ```bash
+   pnpm elevasis-sdk project:list --status active --pretty
+   pnpm elevasis-sdk project:list --status blocked --pretty
+   ```
+
+   Show a compact table: client, status, last-touched.
+
+2. **Surface most-recently-touched project's resume context** — identify the top row (most recent `updated_at`), then print its latest task's `resume_context` (via `project:task:list` → pick most recent in-progress task → `project:task:resume <id> --pretty`). Keep it short: current state + next steps, no files dump.
+
+3. **List available operations** — a one-liner reminder of what the user can do from here:
+
+   ```
+   Available: /project status <client> | /project work <query> | /project task add | /project note | /project create
+   ```
+
+4. **Offer next-action suggestions** — 1-3 concrete follow-ups driven by what's on screen. Examples:
+   - "3 active projects, 1 blocked (Acme — waiting on credentials). Resume Beta (most recent)?"
+   - "No active projects. Create one with `/project create`?"
+   - "Acme's Phase 2 has 2 overdue tasks. Open status?"
+
+### Presentation template
+
+```
+Project Portfolio (2 active, 1 blocked)
+=======================================
+| Client       | Status   | Last touched |
+|--------------|----------|--------------|
+| Beta LLC     | on_track | 2h ago       |
+| Acme Corp    | blocked  | 1d ago       |
+| Gamma Inc    | active   | 4d ago       |
+
+Most recent: Beta LLC — API integration
+  Current: Endpoint wired, tests passing locally.
+  Next:    Deploy to staging, verify webhook delivery.
+
+Next? Resume Beta | Status Acme (why blocked?) | Create new project
+```
+
+---
+
+## Intent Detection (opening-message routing)
+
+Apply this heuristic on the user's first substantive message of a session, before branching into any other operation. The goal: distinguish **resume existing work** from **start new work**, and confirm before committing.
+
+### Signals → classification
+
+| Signal in opening message                                                                       | Classify as                        |
+| ----------------------------------------------------------------------------------------------- | ---------------------------------- |
+| References a specific file path, PR, commit, or existing task/milestone/project by name         | **Resume**                         |
+| "Continue", "pick up", "finish", "the thing we were doing", "yesterday's work"                  | **Resume**                         |
+| "Let's build X", "new Y", "I want to add Z", "start a new project", "help me with a new client" | **New work**                       |
+| Single-file cosmetic ask: "fix typo in X", "rename Y", "update dep Z"                           | **Trivial** (skip new-work prompt) |
+| Ambiguous / discussion only                                                                     | Ask before routing                 |
+
+### Routing
+
+- **Resume** — run `project:work <query>` with the extracted entity (file path stem, project name, task keywords). If one clean match, print the brief and ask "Continue here?". If multiple matches, show top 3 and ask which. If zero matches, fall back to orientation mode with a note ("No match for 'acme webhook'; here's the landscape").
+- **New work** — enter the **New-Work Project Decision** flow (next section).
+- **Trivial** — proceed directly; do not prompt about projects. Small single-file edits don't need portfolio tracking.
+- **Ambiguous** — reply with orientation mode output and ask "Resume something, or start new?"
+
+**Always confirm before branching.** A misclassified "new" routed as "resume" burns trust faster than a one-line confirmation.
+
+---
+
+## New-Work Project Decision
+
+When intent detection classifies the opening ask as **new multi-step work** and the conversation has no established project context, ask this question before writing any code:
+
+```
+Scope this work under a project?
+  a) Create new project "<proposed-slug>"     (recommended for multi-step work)
+  b) Link to existing "<closest-match>"       (if fuzzy-search finds one)
+  c) Standalone — no project tracking         (one-off / trivial)
+```
+
+### Rules
+
+- **Default to (a)** when the ask is multi-step (multiple files, multiple sessions expected, client-facing deliverable).
+- **Offer (b)** only when `project:list` fuzzy-matches produce a plausible candidate (name similarity, shared entity references). If nothing matches, omit the option — do not invent one.
+- **Skip the prompt entirely** for trivial single-file edits — intent detection's "Trivial" branch handles this.
+- **Propose a slug** by extracting the core noun from the ask. "Build a lead scraper for Acme" → `acme-lead-scraper`. User can override.
+
+### After the decision
+
+- **(a) Create** — invoke the guided `/project create` QnA flow (see below), pre-filling the name/description from the ask.
+- **(b) Link** — run `project:resolve <query>` to confirm the match, then add a task to that project for the new work via `project:task:create`.
+- **(c) Standalone** — proceed without project tracking. Do not create orphan task docs; standalone work leaves no artifact.
+
+---
+
+## Connection
+
+### Environment
+
+```bash
+# .env (project root)
+ELEVASIS_PLATFORM_KEY=sk_...
+```
+
+`ELEVASIS_PLATFORM_KEY` is read from the environment automatically. The SDK CLI walks up
+directories to find `.env`, so it resolves correctly from both the project root and
+`operations/`. `ELEVASIS_API_URL` defaults to `https://api.elevasis.io` (production). Set
+`ELEVASIS_API_URL=http://localhost:5170` to target a local API instance, or set
+`NODE_ENV=development` if the SDK CLI respects that convention.
+
+No `--prod` flag is needed — the template targets production by default via `ELEVASIS_API_URL`.
+
+### CLI Invocation
+
+All project operations go through the `elevasis-sdk` CLI. Use the wrapper script form from the
+project root:
+
+```bash
+pnpm elevasis-sdk project:<command> [args]
+```
+
+The long form `pnpm -C operations exec elevasis-sdk project:<command> [args]` is equivalent and
+still works — use it if the wrapper script is not yet available in an older project.
+
+Organization scoping is handled server-side via `ELEVASIS_PLATFORM_KEY`. No org header or
+`--org` flag is required.
+
+---
+
+## Database Schema
+
+### Tables
+
+| Table            | Purpose                      | Key Columns                                                           |
+| ---------------- | ---------------------------- | --------------------------------------------------------------------- |
+| `prj_projects`   | Client projects/contracts    | name, kind, status, description, contract_value, start/end dates      |
+| `prj_milestones` | Phases within projects       | name, status, due_date, sequence, completed_at, checklist             |
+| `prj_tasks`      | Work items within milestones | name, status, type, due_date, milestone_id, resume_context, checklist |
+| `prj_notes`      | Meeting notes, updates       | type, content, summary, occurred_at, task_id, milestone_id            |
+
+### Project Kind
+
+| Kind                | Description                      |
+| ------------------- | -------------------------------- |
+| `client_engagement` | External client delivery project |
+| `internal`          | Internal project work            |
+| `research`          | Research and investigation       |
+| `other`             | Catch-all                        |
+
+Default filter for `/project` operations is `kind=client_engagement` unless the user's intent
+is clearly internal or `--kind` is specified explicitly.
+
+### Status Values
+
+| Entity    | Values                                                                                                                   |
+| --------- | ------------------------------------------------------------------------------------------------------------------------ |
+| Project   | `active`, `on_track`, `at_risk`, `blocked`, `completed`, `paused`                                                        |
+| Milestone | `upcoming`, `in_progress`, `completed`, `overdue`, `blocked`                                                             |
+| Task      | `planned`, `in_progress`, `blocked`, `completed`, `cancelled`, `submitted`, `approved`, `rejected`, `revision_requested` |
+| Note type | `call_note`, `status_update`, `issue`, `blocker`, `agent_learning`                                                       |
+| Task type | `documentation`, `code`, `report`, `design`, `refactor`, `feature`, `bug`, `research`, `other`                           |
+
+### Checklist Shape
+
+Both `prj_milestones.checklist` and `prj_tasks.checklist` store a JSONB array of objects with this shape:
+
+```json
+[
+  { "id": "uuid", "label": "Item label", "completed": false },
+  { "id": "uuid", "label": "Another item", "completed": true }
+]
+```
+
+The CLI `--checklist` flag on both `project:milestone:update` and `project:task:update` performs a **full replace** — the entire array is replaced with the JSON you supply. There are no item-level add/toggle/remove flags. To mutate a single item:
+
+1. Read the current checklist (via psql or `project:task:get` / `project:milestone:list`)
+2. Mutate the array in memory (append, flip `completed`, filter out)
+3. Write the entire array back via `--checklist '<json>'`
+
+To clear a checklist: `--checklist '[]'`.
+
+### Foreign Keys
+
+- `prj_milestones.project_id` → `prj_projects.id` (CASCADE)
+- `prj_tasks.project_id` → `prj_projects.id` (CASCADE)
+- `prj_tasks.milestone_id` → `prj_milestones.id` (SET NULL)
+- `prj_tasks.parent_task_id` → `prj_tasks.id` (SET NULL) — for subtasks
+- `prj_notes.project_id` → `prj_projects.id` (CASCADE)
+- `prj_notes.task_id` → `prj_tasks.id` (SET NULL)
+- `prj_notes.milestone_id` → `prj_milestones.id` (SET NULL)
+- `prj_projects.deal_id` → `acq_deals.id` (SET NULL)
+- `prj_projects.client_id` → `clients.id` (SET NULL) — clients-hub link; set via `--client`
+- `prj_projects.client_company_id` → `acq_companies.id` (SET NULL) — legacy direct-company link; set via `--client-company-id`
+
+---
+
+## Operations
+
+### `list` — Portfolio Table (no orientation extras)
+
+For a plain project table without resume context or next-action suggestions (use the bare `/project` orientation mode above when the user wants the full hub experience):
+
+```bash
+pnpm elevasis-sdk project:list --kind client_engagement --pretty
+
+# Filter by client (accepts the client's name or UUID)
+pnpm elevasis-sdk project:list --client "Acme" --pretty
+```
+
+Present as:
+
+```
+Project Portfolio
+=================
+| Client             | Status   | Kind               | Value   |
+|--------------------|----------|--------------------|---------|
+| Acme Corp          | on_track | client_engagement  | $5,000  |
+| Beta LLC           | active   | client_engagement  | $3,500  |
+```
+
+For progress detail (milestones/tasks), follow up with `project:milestone:list` and
+`project:task:list` per project.
+
+### `work <query>` — Resume Brief
+
+Wraps `project:work` for fuzzy-matched resume. Used by intent-detection "resume" classification.
+
+```bash
+pnpm elevasis-sdk project:work <query>
+```
+
+Prints the matched project/task brief with current `resume_context`. If multiple matches, show top 3 and ask which.
+
+### `status [<client>]` — Detailed Status
+
+If client specified, show full detail for that project. If not, show all.
+
+**Resolve the project first** (see Client Inference below), then:
+
+```bash
+# Get project details
+pnpm elevasis-sdk project:get <project-id>
+
+# Milestones (ordered by sequence)
+pnpm elevasis-sdk project:milestone:list --project <project-id> --pretty
+
+# Tasks (grouped by milestone)
+pnpm elevasis-sdk project:task:list --project <project-id> --pretty
+
+# Recent notes
+pnpm elevasis-sdk project:note:list --project <project-id> --pretty
+```
+
+Present as:
+
+```
+Acme Corp — On Track
+=====================
+Description: Automation project for AI-powered workflows
+Contract: $5,000 | Started: 2026-04-01 | Target: 2026-06-01
+
+Milestones (2/4 complete)
+  [x] Phase 1: Discovery (completed)
+  [>] Phase 2: Implementation (in_progress, due 2026-05-01)
+  [ ] Phase 3: Testing (upcoming, due 2026-05-15)
+  [ ] Phase 4: Launch (upcoming, due 2026-06-01)
+
+Tasks (3/8 approved)
+  Phase 2: Implementation
+    [approved] Requirements doc (documentation)
+    [in_progress] API integration (code, due 2026-05-01)
+    [planned] User guide (documentation)
+
+Recent Notes
+  2026-04-05 [call_note] Weekly check-in — on track, API work started
+  2026-04-01 [call_note] Kickoff — confirmed scope and timeline
+```
+
+#### Tenant Skill Bindings Footer
+
+After the normal status output, append a short footer when the project kind, tags, name,
+description, milestones, or active tasks map to a known tenant domain such as `prospecting`,
+`crm`, `outreach`, `finance`, or `support`.
+
+Resolve the best domain from explicit metadata first (`tags`, `kind`, domain fields), then from
+project/task text. If no clear domain maps, omit the footer.
+
+Footer shape:
+
+```text
+Related skill bindings
+- Domain: <domain>
+- Read or change business profile: /om <domain>
+- Layering primer: node_modules/@elevasis/sdk/reference/spine/spine-primer.md
+```
+
+External projects do not expose monorepo-only architecture commands. Never suggest `/org-os`,
+`/om-spine`, or monorepo paths in this footer.
+
+### `create` — Guided Project Creation (QnA Flow)
+
+Interactive walkthrough that collects project details step by step using `AskUserQuestion`,
+then creates the project, milestones, and kickoff note in one go.
+
+**Step 1: Project basics**
+
+Ask via `AskUserQuestion` (single question, free-text via "Other"):
+
+```
+What's the project name and a brief description of the scope?
+```
+
+The user will reply with something like "Acme Corp automation -- building a lead gen pipeline
+with email outreach". Parse the name and description from the response.
+
+**Step 2: Deal/company linking**
+
+Search for matching records using read-only psql (the psql role is SELECT-only).
+
+Note: `SUPABASE_READONLY_URL` must be set in the template `.env` for psql queries. If not
+set, skip this step and advise the user to link via the Command Center UI after creation.
+
+If `SUPABASE_READONLY_URL` is available, query using `<ORG_ID>` from the platform key's
+associated organization (you can retrieve it via `project:list` output which includes
+`organization_id`):
+
+```bash
+source .env && psql "$SUPABASE_READONLY_URL" -c \
+  "SELECT id, name, domain FROM acq_companies WHERE organization_id = '<ORG_ID>' AND name ILIKE '%<parsed_name>%' ORDER BY updated_at DESC LIMIT 5;"
+
+source .env && psql "$SUPABASE_READONLY_URL" -c \
+  "SELECT d.id, d.contact_email, d.cached_stage, c.name AS company_name
+   FROM acq_deals d
+   LEFT JOIN acq_contacts ct ON ct.id = d.contact_id
+   LEFT JOIN acq_companies c ON c.id = ct.company_id
+   WHERE d.organization_id = '<ORG_ID>'
+     AND (d.contact_email ILIKE '%<parsed_name>%' OR c.name ILIKE '%<parsed_name>%')
+   ORDER BY d.updated_at DESC LIMIT 5;"
+```
+
+If matches found, ask via `AskUserQuestion`:
+
+```
+Found matching records. Link this project?
+- [Company: Acme Corp (acme.com)]
+- [Deal: john@acme.com (closed_won)]
+- No linking
+```
+
+If no matches, skip silently.
+
+**Step 3: Contract details**
+
+Ask via `AskUserQuestion`:
+
+```
+Contract details?
+- Options with previews showing common structures:
+  - "Fixed project" -- one-time fee
+  - "Monthly retainer" -- recurring
+  - "Hourly" -- time & materials
+  - Other
+```
+
+Then follow up for the value, start date, and target end date. These can be collected in a
+single question:
+
+```
+What's the contract value, start date, and target end date?
+(e.g., "$5000, starting today, targeting June 1")
+```
+
+Parse natural language dates ("today", "next Monday", "June 1") into ISO format.
+
+**Step 4: Milestone suggestions**
+
+Based on the project description from Step 1, suggest 3-5 milestones that fit the scope. Use
+your understanding of the project to propose relevant phases.
+
+**Guidelines for milestone suggestions:**
+
+- For automation/integration projects: Discovery, Implementation, Testing, Launch
+- For content/marketing projects: Strategy, Content Creation, Review, Distribution
+- For consulting/advisory: Assessment, Recommendations, Implementation Support
+- Always include a Discovery/Kickoff phase first and a Launch/Handoff phase last
+- Suggest due dates spaced evenly between start and target end date
+
+Present via `AskUserQuestion` with previews:
+
+```
+Suggested milestones for "<project name>". Select which to include:
+(multiSelect: true)
+
+- Discovery (due <start + 2 weeks>)
+- Implementation (due <midpoint>)
+- Testing (due <end - 2 weeks>)
+- Launch (due <end date>)
+```
+
+The user can select which ones to keep, modify via "Other", or skip entirely.
+
+**Step 5: Kickoff note**
+
+Ask via `AskUserQuestion`:
+
+```
+Add a kickoff note? This captures initial context (scope confirmed, key contacts, etc.)
+- Yes, let me type one
+- Skip for now
+```
+
+If yes, collect the note content.
+
+**Step 6: Execute**
+
+Create everything in sequence using the CLI:
+
+```bash
+# 1. Create project
+# If the user mentioned a client by name, resolve it first:
+#   pnpm elevasis-sdk client:resolve "<client name>"
+# Then pass the name (or UUID) via --client
+pnpm elevasis-sdk project:create \
+  --name "<name>" \
+  --kind client_engagement \
+  --status active \
+  --description "<desc>" \
+  --client "<client-name-or-uuid>"
+
+# 2. Create milestones
+pnpm elevasis-sdk project:milestone:create \
+  --project <project-id> \
+  --name "<milestone-name>" \
+  --status upcoming \
+  --due-date <date>
+
+# 3. Create kickoff note (if provided)
+pnpm elevasis-sdk project:note:create \
+  --project <project-id> \
+  --content "<note-content>" \
+  --type call_note
+```
+
+If `--company` or `--deal` linking was confirmed, use `project:update` after creation:
+
+```bash
+pnpm elevasis-sdk project:update <project-id> --client "<client-name-or-uuid>"
+```
+
+To remove a client link later:
+
+```bash
+pnpm elevasis-sdk project:update <project-id> --clear-client
+```
+
+Do NOT pass `--client ""` to clear — an empty string is rejected as ambiguous. Use `--clear-client` instead.
+
+**Step 7: Summary**
+
+Show the created project:
+
+```
+Project Created
+===============
+Name: Acme Corp Automation
+Kind: client_engagement
+Status: active
+Timeline: 2026-04-07 → 2026-06-01
+
+Milestones:
+  1. Discovery (upcoming, due 2026-04-21)
+  2. Implementation (upcoming, due 2026-05-08)
+  3. Testing (upcoming, due 2026-05-22)
+  4. Launch (upcoming, due 2026-06-01)
+
+Kickoff Note: Confirmed scope -- lead gen pipeline with email outreach.
+
+Next: /project status acme
+```
+
+---
+
+### `add "<name>" [options]` — Quick Create Project (No QnA)
+
+**Options:**
+
+- `--status <status>` (default: `active`)
+- `--value <number>` — contract value (informational, stored in description or notes)
+- `--start <date>` — start date
+- `--end <date>` — target end date
+- `--description "<text>"` — scope description
+- `--kind <kind>` (default: `client_engagement`)
+
+```bash
+pnpm elevasis-sdk project:create \
+  --name "<name>" \
+  --kind client_engagement \
+  --status active \
+  --description "<desc>"
+```
+
+**After creating:** Show the created project and suggest next steps (add milestones).
+
+### `update <client> [options]` — Update Project
+
+**Options:** `--status`, `--name`, `--description`, `--client <id-or-name>`, `--clear-client`, `--client-company-id <uuid>`
+
+First resolve the client (see Client Inference below), then:
+
+```bash
+pnpm elevasis-sdk project:update <project-id> --status <status>
+```
+
+### `milestone add <client> "<name>" [options]` — Add Milestone
+
+**Options:**
+
+- `--status <status>` (default: `upcoming`)
+- `--due <date>` — due date (ISO)
+- `--description "<text>"`
+
+```bash
+pnpm elevasis-sdk project:milestone:create \
+  --project <project-id> \
+  --name "<name>" \
+  --status upcoming \
+  --due-date <date>
+```
+
+### `milestone update <client> "<milestone>" [options]` — Update Milestone
+
+Resolve milestone by name (see Client Inference for project resolution, then use
+`project:milestone:list` to find the milestone ID by partial name match):
+
+```bash
+# List milestones to find ID
+pnpm elevasis-sdk project:milestone:list --project <project-id>
+
+# Update status
+pnpm elevasis-sdk project:milestone:update <milestone-id> --status completed
+
+# Update checklist (full replace)
+pnpm elevasis-sdk project:milestone:update <milestone-id> \
+  --checklist '[{"id":"uuid","label":"Item label","completed":false}]'
+```
+
+If status changes to `completed`, the API auto-sets `completed_at`.
+
+**Options:** `--status`, `--due-date`, `--name`, `--checklist <json>`
+
+**Checklist mutations:** The CLI performs a full replace. To add, toggle, or remove an item:
+
+1. Read the current checklist via psql or `project:milestone:list`
+2. Mutate the array in memory
+3. Write back the full array via `--checklist '<json>'`
+
+### `checklist <client> "<milestone>"` — View Checklist
+
+Resolve the milestone, then read its `checklist` JSONB column via read-only psql.
+
+Note: Requires `SUPABASE_READONLY_URL` in the template `.env`. If not set, advise the user
+to view the checklist via the Command Center UI, or add `SUPABASE_READONLY_URL` to `.env`.
+
+```bash
+source .env && psql "$SUPABASE_READONLY_URL" -c \
+  "SELECT m.name, m.checklist
+   FROM prj_milestones m
+   WHERE m.project_id = '<project_id>' AND m.name ILIKE '%<name>%'
+   LIMIT 1;"
+```
+
+The `checklist` column is a JSONB array of objects:
+
+```json
+[
+  { "id": "uuid", "label": "Set up dev environment", "completed": true },
+  { "id": "uuid", "label": "Draft scope document", "completed": false }
+]
+```
+
+Present as:
+
+```
+Onboarding & Scope — Checklist (2/3 complete)
+  [x] Set up dev environment
+  [x] Schedule kickoff call
+  [ ] Draft scope document
+```
+
+### `task add <client> "<name>" [options]` — Add Task
+
+**Options:**
+
+- `--milestone "<name>"` — link to a milestone (resolve by name first)
+- `--status <status>` (default: `planned`)
+- `--type <type>` (default: `other`)
+- `--due <date>`
+- `--description "<text>"`
+- `--checklist <json>` — initial checklist items
+
+```bash
+# Resolve milestone ID first if --milestone provided
+pnpm elevasis-sdk project:milestone:list --project <project-id>
+
+# Create task
+pnpm elevasis-sdk project:task:create \
+  --project <project-id> \
+  --title "<name>" \
+  --status planned \
+  --type <type> \
+  --milestone <milestone-id>
+
+# Create task with initial checklist
+pnpm elevasis-sdk project:task:create \
+  --project <project-id> \
+  --title "<name>" \
+  --checklist '[{"id":"1","label":"Step one","completed":false}]'
+```
+
+### `task update <client> "<task>" [options]` — Update Task
+
+Resolve task by name within the project using `project:task:list`, then update by positional ID:
+
+```bash
+# Find the task ID
+pnpm elevasis-sdk project:task:list --project <project-id>
+
+# Update status (positional <id> comes first, then flags)
+pnpm elevasis-sdk project:task:update <task-id> --status <status>
+
+# Update checklist (full replace)
+pnpm elevasis-sdk project:task:update <task-id> \
+  --checklist '[{"id":"uuid","label":"Step","completed":false}]'
+
+# Clear checklist
+pnpm elevasis-sdk project:task:update <task-id> --checklist '[]'
+```
+
+**Options:** `--status`, `--title`, `--milestone`, `--description`, `--checklist <json>`
+
+**Important:** The task ID is a **positional argument** — it comes immediately after `project:task:update`, before any flags. There is no `--task-id` flag.
+
+If status changes to `approved`, the API auto-sets `completed_at`.
+
+**Checklist mutations:** Same full-replace semantics as milestones. To add, toggle, or remove an item:
+
+1. Read the current task via `project:task:get <task-id>`
+2. Mutate the checklist array in memory
+3. Write back the full array via `project:task:update <task-id> --checklist '<json>'`
+
+### `note <client> "<content>" [options]` — Add Note
+
+**Options:**
+
+- `--type <type>` (default: `call_note`) — see Note Type Selection below
+- `--task <task-id>` — attach to a specific task (UUID)
+- `--milestone <milestone-id>` — attach to a milestone (UUID)
+
+```bash
+pnpm elevasis-sdk project:note:create \
+  --project <project-id> \
+  --content "<content>" \
+  --type <type>
+
+# Attach to a task
+pnpm elevasis-sdk project:note:create \
+  --project <project-id> \
+  --content "<content>" \
+  --type agent_learning \
+  --task <task-id>
+```
+
+### `notes <client>` — List Notes
+
+```bash
+pnpm elevasis-sdk project:note:list --project <project-id> --pretty
+```
+
+### `delete <client>` — Delete Project
+
+**ALWAYS confirm before deleting.** Show what will be deleted (milestone count, task count,
+note count) using the list commands, then ask for confirmation.
+
+```bash
+pnpm elevasis-sdk project:delete <project-id>
+```
+
+Cascade deletes all milestones, tasks, and notes.
+
+---
+
+## Notes Management
+
+Notes (`prj_notes`) are the durable record of project-scoped signals — conversation outcomes, blockers, status updates, and agent-captured learnings. They complement `resume_context` (which is task-scoped, temporal, and agent-facing) by providing a human-readable audit trail.
+
+### When to Create a Note vs. Fall Through to Global Auto-Memory
+
+**Create a project note** when all of the following hold:
+
+- A specific project and (optionally) task is in scope in the current session
+- The content is project-bound — it would only matter to someone working on this project
+- The content is durable — it should be visible in future status views, not just the current session
+
+**Fall through to global auto-memory** when:
+
+- No active project or task can be resolved from session context (see Project-Scope Resolution below)
+- The content is cross-project or platform-wide (e.g., a general TypeScript pattern, a reusable workflow insight)
+- The signal is ephemeral and not worth a DB write (e.g., "reminder: run linter before committing")
+
+When in doubt and no project context is resolvable, global auto-memory is the safe default.
+
+### Note Type Selection Guide
+
+| Type             | Use when                                                                                                                                                                                                                                                   |
+| ---------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `call_note`      | Transcribing or summarizing a live client / stakeholder call. Include key decisions and action items.                                                                                                                                                      |
+| `status_update`  | Milestone-level progress worth flagging to a human operator: phase complete, substantial deliverable, direction change. Don't emit one every session turn.                                                                                                 |
+| `issue`          | Bug, regression, unexpected failure, contract mismatch. Include reproduction context.                                                                                                                                                                      |
+| `blocker`        | Work cannot proceed without external action: decision needed, credential missing, upstream fix required.                                                                                                                                                   |
+| `agent_learning` | Project-bound knowledge the agent discovered: API quirks, client conventions, deploy gotchas, rate limits. Captures "things a future agent resuming this project should know" so it travels with the project instead of living only in global auto-memory. |
+
+### Project-Scope Resolution
+
+Before writing a note, the agent resolves which project (and optionally task) to attach it to:
+
+1. **Explicit in session** — a project/task UUID was mentioned or returned by a recent `project:work <id>` or `project:task:*` invocation in this session. Use it directly.
+2. **Implicit from task list** — if step 1 yields nothing, run (`--status` accepts one value; run both calls and merge):
+
+   ```bash
+   pnpm elevasis-sdk project:list --status active --pretty
+   pnpm elevasis-sdk project:list --status blocked --pretty
+   pnpm elevasis-sdk project:task:list --project <top-project-id> --status in_progress
+   ```
+
+   Use the most-recently-touched project's in-progress task if there is exactly one. If multiple in-progress tasks exist, prefer the one that most closely matches the current conversation topic.
+
+3. **Prompt the user** — if steps 1 and 2 are both ambiguous or empty, ask once: "Which project should I attach this note to? (project UUID or slug, or 'skip' to use global auto-memory)."
+
+4. **Fall through** — if the user says "skip" or no resolution is possible, write to global auto-memory and continue.
+
+Resolution is **agent-side at call time** — there is no middleware or automatic routing layer. The agent infers context from the session.
+
+### Agent Workflow: Capturing an `agent_learning`
+
+When the agent recognizes a project-scoped learning (an API quirk, an undocumented constraint, a client convention discovered during implementation):
+
+1. Recognize the signal — e.g., "Apify actor X rate-limits at 20 rps", "client requires ISO dates, not timestamps", "staging env needs VPN before webhook delivery works"
+2. Resolve active project and task from session context (see above)
+3. Write the note:
+
+   ```bash
+   pnpm elevasis-sdk project:note:create \
+     --project <project-id> \
+     --type agent_learning \
+     --task <task-id> \
+     --content "<concise, factual statement of what was learned>"
+   ```
+
+   Omit `--task` if no specific task is in scope (e.g., the learning is milestone-level or project-level).
+
+4. If no project can be resolved, write the learning to global auto-memory instead.
+
+**Content style for `agent_learning` notes:** write as a factual assertion another agent could act on — "Apify actor X rate-limits at 20 rps; add a 50ms delay between calls" rather than "I noticed that Apify seems slow." Keep it under 200 characters when possible.
+
+---
+
+## Status Transitions from Other Skills
+
+The `/project` skill is the canonical reference for how lifecycle signals from other skills map to `project:task:update` status transitions. The behavioral hooks live in the emitting skills; this section documents the contract.
+
+### Transition Map
+
+| Signal                                   | Source skill   | Transition                             | CLI call                                           |
+| ---------------------------------------- | -------------- | -------------------------------------- | -------------------------------------------------- |
+| Successful `git push` via `/deploy`      | `deploy/`      | linked task → `submitted`              | `project:task:update <task-id> --status submitted` |
+| `/save` catches "I'm stuck" / blocker    | `save/`        | linked task → `blocked` + blocker note | `project:task:update <task-id> --status blocked`   |
+| User/agent says "I'm done" / "task done" | `/project` NLM | active task → `completed`              | `project:task:update <task-id> --status completed` |
+
+### "I'm Done" Recognition (Natural Language Mode)
+
+When the user or agent says any of the following in the context of a project task, treat it as a `completed` transition signal:
+
+- "I'm done", "I'm finished", "done with this", "that's done"
+- "task complete", "task finished", "mark as complete"
+- "wrap this up", "close this task", "finalize this"
+
+Resolve the active task from session context (see Project-Scope Resolution), confirm once ("Mark `<task name>` as completed?"), then fire:
+
+```bash
+pnpm elevasis-sdk project:task:update <task-id> --status completed
+```
+
+If no task resolves, ask the user which task to complete rather than silently skipping.
+
+### Transition Rules (All Skills)
+
+These rules apply to every skill that emits a transition:
+
+- **Explicit task ID only** — use the UUID resolved from session context. Never guess.
+- **Best-effort** — if the CLI returns non-2xx, emit a warning line and continue. Do not block the primary operation.
+- **Idempotent** — if the task is already in the target status, the update is a no-op. Fire it anyway.
+- **Confirm before `completed`** — always ask once before marking a task done, since this is hard to undo semantically.
+
+For full implementation details, see the emitting skills: `deploy/SKILL.md` (Step 8) and `save/SKILL.md` (Steps 5, 5a).
+
+---
+
+## Client Inference
+
+Resolve which project the user means using these rules (in priority order):
+
+1. **Exact ID** — if the argument looks like a UUID, use it directly with `project:get`
+2. **Name match** — list projects filtered by `client_engagement` kind, then match by name:
+   ```bash
+   pnpm elevasis-sdk project:list --kind client_engagement
+   ```
+   Filter results by partial name match against the user's input (case-insensitive).
+3. **Single active project** — if only one non-completed `client_engagement` project exists,
+   use it without asking
+4. **Context from conversation** — if the conversation has been discussing a specific client,
+   use that one
+5. **Ambiguous** — if multiple match or none match, list projects and ask
+
+---
+
+## Natural Language Mode
+
+When args don't match any command pattern, infer intent from natural language:
+
+| User says                                                      | Inferred operation                                                                                                                                    |
+| -------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- |
+| "how's acme doing"                                             | `status acme`                                                                                                                                         |
+| "add a note: weekly call, everything on track"                 | `note <active-project> "weekly call, everything on track"`                                                                                            |
+| "mark phase 2 complete for acme"                               | `milestone update acme "phase 2" --status completed`                                                                                                  |
+| "approve the API docs task"                                    | `task update <project> "API docs" --status approved`                                                                                                  |
+| "new client: Beta LLC, $3500/mo, starting may 1"               | `add "Beta LLC" --value 3500 --start 2026-05-01`                                                                                                      |
+| "what's overdue?"                                              | List milestones where `due_date < now() AND status != 'completed'`                                                                                    |
+| "block acme, waiting on client credentials"                    | `update acme --status blocked`                                                                                                                        |
+| "show checklist for phase 2"                                   | Read and display checklist for milestone "phase 2" via psql or `project:milestone:list`                                                               |
+| "add 'deploy staging' to task X's checklist"                   | Read task X's checklist → append `{id: uuid, label: "deploy staging", completed: false}` → `project:task:update <id> --checklist '<json>'`            |
+| "mark 'deploy staging' done on task X"                         | Read task X's checklist → flip `completed` on matching item → `project:task:update <id> --checklist '<json>'`                                         |
+| "clear the checklist on task X"                                | `project:task:update <task-id> --checklist '[]'`                                                                                                      |
+| "add checklist item to onboarding milestone: review scope doc" | Read milestone checklist → append item → `project:milestone:update <id> --checklist '<json>'`                                                         |
+| "I'm done" / "task complete" / "done with this"                | Resolve active task → confirm → `project:task:update <task-id> --status completed`                                                                    |
+| "save: Apify actor X rate-limits at 20rps"                     | Resolve active project/task → `project:note:create --project <id> --type agent_learning --task <task-id> "Apify actor X rate-limits at 20rps"`        |
+| "remember: client requires ISO dates, not timestamps"          | Resolve active project/task → `project:note:create --project <id> --type agent_learning --task <task-id> "client requires ISO dates, not timestamps"` |
+
+**Checklist note:** all checklist mutations use the read-modify-write pattern. The CLI has no item-level flags (`--add-item`, `--toggle`, `--remove-item` do not exist). Always read the current state, mutate the array, then write the full array back.
+
+---
+
+## Bulk Operations
+
+### Add Multiple Milestones
+
+For creating a standard milestone set, run `project:milestone:create` in sequence:
+
+```bash
+for name in "Discovery" "Implementation" "Testing" "Launch"; do
+  pnpm elevasis-sdk project:milestone:create \
+    --project <project-id> \
+    --name "$name" \
+    --status upcoming
+done
+```
+
+### Overdue Report
+
+Query via read-only psql for milestones past their due date. Requires `SUPABASE_READONLY_URL`
+in the template `.env`. Replace `<ORG_ID>` with the organization ID associated with your
+`ELEVASIS_PLATFORM_KEY` (visible in `project:list` output as `organization_id`). If
+`SUPABASE_READONLY_URL` is not configured, use `project:milestone:list` per project and filter
+manually.
+
+```bash
+source .env && psql "$SUPABASE_READONLY_URL" -c \
+  "SELECT
+     p.name AS project,
+     m.name AS milestone,
+     m.due_date,
+     m.status,
+     (CURRENT_DATE - m.due_date::date) AS days_overdue
+   FROM prj_milestones m
+   JOIN prj_projects p ON p.id = m.project_id
+   WHERE p.organization_id = '<ORG_ID>'
+     AND m.due_date < CURRENT_DATE
+     AND m.status NOT IN ('completed')
+   ORDER BY m.due_date;"
+```
+
+---
+
+## Client Linking
+
+Projects can be linked to a client in the clients hub. When a user mentions a client by name, resolve it to an ID and use `--client` to wire the project.
+
+### Two linking flags, two different columns
+
+- `--client <name-or-uuid>` — links to a clients-hub record (`prj_projects.client_id`). Use this for new client linkage. Accepts a name (the CLI fuzzy-resolves it) or a UUID.
+- `--client-company-id <uuid>` — legacy direct-company linkage (`prj_projects.client_company_id`); kept for back-compat. UUID only.
+
+These are independent. Do not use `--client-company-id` when you mean `--client`.
+
+### When the user mentions a client by name
+
+1. Resolve the client name to a UUID first (optional but gives an explicit confirmation step):
+   ```bash
+   pnpm elevasis-sdk client:resolve "Acme"
+   ```
+2. Pass the name or UUID to `--client` — the CLI fuzzy-resolves names automatically:
+   ```bash
+   pnpm elevasis-sdk project:create --name "Acme Automation" --kind client_engagement --client "Acme"
+   pnpm elevasis-sdk project:update <project-id> --client "Acme"
+   ```
+3. If the client name matches multiple records, the CLI returns an error listing candidates. In that case, run `client:resolve "Acme"` to disambiguate, then pass the UUID directly.
+
+### Filtering projects by client
+
+```bash
+pnpm elevasis-sdk project:list --client "Acme" --pretty
+```
+
+### Removing a client link
+
+```bash
+pnpm elevasis-sdk project:update <project-id> --clear-client
+```
+
+`--client` and `--clear-client` are mutually exclusive on the same command call.
+
+### Soft recommendation for client engagement projects
+
+When the user creates a `client_engagement` project without mentioning a client, gently note that linking a client is recommended so the project appears in client lineage views. Do not block creation — the flag is optional.
+
+### Client command reference
+
+The full `client:*` surface (list, get, status, resolve) is available via `elevasis-sdk client:*`. The `client:resolve` command mirrors `project:resolve` in shape and is the canonical tool for name-to-ID translation.
+
+---
+
+## Safety Rules
+
+1. **Always confirm deletes** — show what will be cascade-deleted before executing
+2. **Organization scoping** — all API calls are automatically org-scoped via
+   `ELEVASIS_PLATFORM_KEY`; every direct psql query MUST include an `organization_id` filter
+   (use `<ORG_ID>` as a placeholder and confirm the value with the user if unknown)
+3. **Read before write** — when updating, show current state before applying changes
+4. **Validate status values** — reject invalid status strings before sending to the CLI
+5. **Kind default** — default to `--kind client_engagement` when user intent is client
+   delivery; use `--kind internal` for internal project work
+6. **Template project assumption** — the template is a single-organization project. All
+   `/project` operations operate within the organization scoped by `ELEVASIS_PLATFORM_KEY`.
+   There is no cross-org capability.
+7. **Checklist full-replace** — the `--checklist` flag replaces the entire array. Always read
+   the current checklist before writing to avoid losing existing items.
+8. **Confirm task completion** — always ask once before marking a task `completed` via "I'm done"
+   recognition. This status is semantically significant and warrants a confirmation step.
+
+---
+
+**Last Updated:** 2026-05-08