recall · diff
git:20260508.65c1b0b to git:20260618.ba67ec2
40 added, 79 removed. Audit A to A.
---
name: recall
- description: "Spawns a subagent to read a project's save-state files and output a tight 6-field briefing — fully autonomous, no user interaction. Trigger when the user invokes /recall [project-slug], says 'resume session', 'pick up where we left off', or 'continue from last session'. The subagent reads next-session.md, heartbeat.md, decisions.md, and save-state-state.json — all read-only. Outputs: Phase, Next action (1 sentence), Blockers (1 per line), Top 2 locked decisions, 1–2 mid-flight files, and context from the last session. Also for: quickly orienting to an unfamiliar project on first contact, checking whether a project has active work mid-session without reading all files manually, and resuming multi-project context across sessions by running /recall for each slug in sequence. SSOT for project paths is ~/.claude/memory/medium-term.md."
+ description: >
+ Quick single-project context check. Reads next-session.md directly (no subagent)
+ and outputs a 6-field briefing. For multi-project resume, use /pd-resume instead.
+ Invoke as /recall [slug] or when user says 'resume session', 'pick up where we left off'.
+ SSOT for project paths is ~/.claude/memory/medium-term.md.
---
# Recall
- Spawns a subagent that reads the project's save-state files and outputs a briefing.
- Caller spawns and waits — zero work done in the calling session.
-
- ## SSOT: medium-term.md
-
- The **primary source of truth** for project locations is `~/.claude/memory/medium-term.md`
- — the Active Projects table. Read it FIRST.
-
- ## Project Map
-
- Read from `~/.claude/memory/medium-term.md` — this is the SSOT for project paths.
+ Direct-read briefing for one project. No subagents, no temp files.
+ For multi-project autonomous resume, use `/pd-resume` instead.
- | Project | Path |
- |---------|------|
- | {project-a} | `{project-a-directory}` |
- | {project-b} | `{project-b-directory}` |
+ ## SSOT
- (Populate from your medium-term.md Active Projects table.)
+ Project registry: `~/.claude/memory/medium-term.md` — Active Projects table.
+ Read it FIRST. This is the ONLY project-to-path map.
## If Slug Not Found
- 1. Check `~/.claude/memory/medium-term.md` Active Projects table — the SSOT.
- 2. If not in medium-term.md → output:
-
- ```
- PROJECT NOT FOUND: {slug}
- Hint: Check ~/.claude/memory/medium-term.md for the current project list.
- ```
-
- **Stop. Do not fall back to other projects or search broadly.**
-
- ## Step 1 — Spawn Subagent
-
- Use the Agent tool to spawn a general-purpose sonnet subagent. The subagent reads files
- and outputs the briefing — do not do any reading yourself.
-
- Subagent prompt:
+ 1. Check `~/.claude/memory/medium-term.md` Active Projects table.
+ 2. If not found → output: `PROJECT NOT FOUND: {slug}` and stop.
- "Read-only briefing. Do NOT write any files.
+ ## Step 1 — Resolve Project Path
- PERMISSIONS: read-only on ALL paths below. No write/edit/create.
+ 1. Read `~/.claude/memory/medium-term.md`
+ 2. Find the slug in the Active Projects table → get project path
+ 3. If not found → stop with error
- ## Scope
- - Memory dir: `{project}/memory/` — next-session.md, heartbeat.md, decisions.md, sessions/
- - State tracking: `{project}/.claude/save-state-state.json`
- - Optional project root files: `{project}/STATE.md`, `{project}/CLAUDE.md`
+ ## Step 2 — Read Briefing
- FILES TO READ:
- - {project}/memory/next-session.md — primary briefing (status, next, blockers, mid-flight)
- - {project}/memory/heartbeat.md — phase status + blockers (not Session End history)
- - {project}/memory/decisions.md — top 2 locked decisions
- - {project}/.claude/save-state-state.json (if exists)
+ Read `{project-path}/memory/next-session.md` directly using the Read tool.
+ **Do NOT spawn a subagent for this.**
- OUTPUT FORMAT (output exactly this, nothing else):
+ next-session.md is self-contained — save-state writes phase, next action,
+ blockers, decisions, mid-flight, and delegated statuses into it.
+ If the file doesn't exist or is empty, output:
+ ```
RECALL — {slug}
-
- Phase: [current phase or status]
- Next: [specific action — one sentence]
- Blockers: [list one per line, or "none"]
- Decisions: [top 2 locked decisions, one per line, or "none"]
- Mid-flight: [1-2 mid-flight files, one per line, or "none"]
- Context: [what was happening last session — 1-2 sentences]
-
- --- Last Session (from heartbeat Session End) ---
- Completed: [from heartbeat "Completed This Session" — top 2-3 bullets]
- Remaining: [from heartbeat "In Progress" + "Top 3 Priorities"]
-
- RULES:
- - Read-only. Do NOT write or edit any files.
- - If a file doesn't exist, skip that field and note "N/A".
- - Keep every field to 1-2 lines max.
- - Be specific — "fix BottomNav.tsx mobile layout" not "fix bugs".
- - Output ONLY the briefing above. No preamble, no commentary."
-
- subagent_type: Explore
- model: sonnet
-
- Wait for the subagent to complete.
+ Phase: unknown
+ Next: read project memory and assess current state
+ Blockers: none
+ Decisions: none
+ Mid-flight: none
+ Context: no prior session found
+ ```
- ## Step 2 — Graph enrichment (caller-side)
+ ## Step 3 — Output Briefing
- After the subagent returns the recall briefing, call:
+ Format the content from next-session.md into:
```
- mcp__graphify__query_graph(question="{slug} key components and active connections")
- ```
-
- Append the top 5 returned node labels as a **Graph:** line at the end of the recall output:
+ RECALL — {slug}
- ```
- Graph: <node1>, <node2>, <node3>, <node4>, <node5>
+ Phase: [from next-session]
+ Next: [from next-session]
+ Blockers: [from next-session, or "none"]
+ Decisions: [from next-session, or "none"]
+ Mid-flight: [from next-session, or "none"]
+ Context: [from next-session status/notes, 1-2 sentences]
```
- If the graphify MCP tool returns an error, append:
- Graph: ⚠️ graphify MCP unavailable — run: graphify serve ~/.claude/graphify-out/unified/graph.json
- Do not silently skip. The graph context is load-bearing for session quality.
+ Output only the briefing. No preamble, no commentary, no follow-up questions.