/arstart

description: "AgentRecall session opener — status board (no args), project loader (<slug>), or cold-start bootstrap. Use first every session."

/arstart — AgentRecall Session Opener

One command, three modes — the single entry point for every AgentRecall session.

| Invocation | Mode | Use when | |---|---|---| | `/arstart` (no args) | **Board** | You don't know what to work on yet — see everything in flight | | `/arstart <slug>` | **Load** | You already know the project — load its full context | | `/arstart bootstrap` | **Bootstrap** | First install, or backfilling projects AgentRecall doesn't know about yet |

**USE THIS FIRST.** Every new session — before any work, before picking up a task.

---

Mode: Board (no arguments)

The true cold start. One command to see everything in flight across all projects — for both humans choosing what to work on and agents loading briefing context.

When to Use

**USE THIS FIRST.** Every new session — before any work.

• Opening a new Claude Code tab or starting a fresh agent
• "What was I working on?" after any break
• Orchestrator loading full briefing before dispatching executor agents
• Deciding which project to pick up next

Board mode only tells you what to pick. Once you know the project, switch to Load mode (`/arstart <slug>`).

What This Does

Scans every project in `~/.agent-recall/projects/`, reads the latest journal's `## Next` section per project, classifies status, and renders a unified status board card.

Process

Step 1: Scan all project directories

ls ~/.agent-recall/projects/

For each subdirectory (project slug), find the latest journal file:

ls ~/.agent-recall/projects/<slug>/journal/*.md 2>/dev/null \
  | grep -v '\-log\.' \
  | grep -v 'index\.md' \
  | sort -r \
  | head -1

Skip any slug that returns no journal files — it's an empty or test directory.

Step 2: Extract date, Next section, and Intention

From each journal filename, extract the date (`YYYY-MM-DD` prefix).

Read the file and extract the content under `## Next` — stop at the next `##` heading or EOF. If no `## Next` section exists, use the first non-frontmatter line of the file as fallback.

Also read the project's intention from its identity file:

cat ~/.agent-recall/projects/<slug>/palace/identity.md 2>/dev/null

Look for a line starting with `**Intention:**` or `Intention:`. Extract the value after the colon. This is the user's original WHY for starting the project — distinct from what's next to do.

Step 3: Classify project status

| Indicator | Status | Condition | |-----------|--------|-----------| | `●` | Active | Journal within 7 days AND has a Next item | | `⚠` | Blocked | Next section contains "Blocked", "blocked on", or "waiting for" | | `✓` | Complete | Next section contains "feature-complete", "shipped", "done", or "complete" | | `-` | Stale | Latest journal older than 14 days |

When a project matches both `⚠` and `●`, show `⚠` (blocked takes priority).

Step 4: Render the status board

Sort order: `⚠` blocked first (needs attention), then `●` active (most recent first), then `✓` complete, then `-` stale.

Special rule: the global/catchall project goes last — it's less actionable than named project slugs.

──────────────────────────────────────────────────────────────
  AgentRecall  Status Board        <YYYY-MM-DD>    <N> projects
──────────────────────────────────────────────────────────────

  1  ⚠ <project-slug>        <YYYY-MM-DD>   BLOCKED
       Why: <intention>  |  Blocked: <blocked reason>

  2  ● <project-slug>        <YYYY-MM-DD>
       Why: <intention>  |  Next: <Next item — max 60 chars>

  3  ● <project-slug>        <YYYY-MM-DD>
       Next: <Next item — one line, max 80 chars>

  4  ✓ <project-slug>        <YYYY-MM-DD>   complete
       <Status note — one line>

  5  - <project-slug>        <YYYY-MM-DD>   stale
       Last: <last entry summary — one line>

──────────────────────────────────────────────────────────────
  Enter a number, or:
    N  New project (with memory — agent knows your full history)
    X  New project (clean slate — no prior context, pure objectivity)
    d<N>  Delete project at that number (e.g. d5)
──────────────────────────────────────────────────────────────

Rules for the card:

• Each project is exactly 2 lines: status line (with number) + content line (indented to align under slug)
• Content line format (priority order):
• If intention exists: prefix with `Why: <intention> | ` then append `Next:` or `Blocked:` or `Last:`
• If no intention: show `Next:` / `Blocked:` / `Last:` alone
• Truncate the whole content line to ~90 chars with `…`
• Numbers start at 1 and increment continuously across all status groups
• `<N> projects` = total count shown (excluding skipped empties)
• Date column aligns across all rows for readability
• The slug shown in the card is for reference only — the human responds with a number, not the slug name

Step 5: Respond to selection (interactive sessions only)

Four response types:

**Number (e.g. `3`)** — existing project Map the number back to the slug from your rendered list. Run `/arstart <slug>` (Load mode) to load full context.

**N — New project with memory** Ask: "Project name?" → create a new project slug (kebab-case, auto-derived from name). Call `session_start(project="<new-slug>")` — this loads cross-project insights and awareness from ALL existing projects. Good for work that builds on or connects to existing projects.

> Intention is captured automatically on first `/arsave` — extracted from the earliest user messages in that conversation. No need to ask here.

**X — New project, clean slate** Ask: "Project name?" → note the slug. Do NOT call session_start. Do NOT load any memory, awareness, or past insights. Say: "Starting fresh — no prior context loaded. This session is objective." Good for: code reviewers, audits, independent evaluations, second opinions.

> Intention is