pmm-recall

name	pmm:recall
description	Context-switching recall — synthesizes a focused briefing on a topic from across all memory files. Use when switching workstreams or coming back to a topic.
argument-hint	[topic]

pmm:recall

Context-switching primitive. When you're jumping between workstreams or coming back to a topic, pmm:recall reads across memory files and synthesizes a focused briefing — not raw search results, but working context you can act on immediately.

Topic: $ARGUMENTS

---

Routing

No arguments — session resume

If $ARGUMENTS is empty, produce a quick resume card:

1. Read memory/last.md and memory/progress.md
2. Synthesize a "where were we?" card:

## Session Resume

**Last**: [1-sentence summary from last.md — what happened most recently]
**State**: [current state from progress.md — active work, blockers]
**Next**: [top 2-3 items from progress.md next section]

Keep it under 10 lines. Ready to work from, not to read.

With topic — focused recall

If $ARGUMENTS contains a topic, synthesize a focused briefing across all relevant files.

---

Execution (topic mode)

Step 1 — Read config

Read memory/config.md. Extract:

• Session Start → Mode and bootstrap_wired
• Active Files — which files are currently active

Step 2 — Context-first routing

If Mode: lazy AND bootstrap_wired: true — memory files are already in context. Execute Steps 3–4 directly in the main context window without dispatching any agent. Tier 1 files are in-context. For Tier 2 files (graph.md, vectors.md, taxonomies.md, assets.md), use the Read tool to load the relevant file only if the topic likely involves relationships, entities, or classifications.

If Mode: eager OR bootstrap_wired: false — fall through to Agent Dispatch at the end of this document.

Step 3 — Search across active files

Search all active Tier 1 files for entries related to the topic (case-insensitive keyword match). For each file, collect relevant entries:

File	What to look for
timeline.md	Recent activity on this topic
last.md	Whether this topic was part of the last session
decisions.md	Key decisions made about this topic, with rationale
lessons.md	Mistakes or lessons related to this topic
progress.md	Current state, blockers, next steps for this area
processes.md	Active workflows or procedures for this topic
standinginstructions.md	Standing rules that apply to this topic
memory.md	Durable facts and context
preferences.md	User preferences relevant to this area
voices.md	Tone or reasoning lenses relevant to this area
summaries.md	Past session summaries mentioning this topic

If Tier 2 files might be relevant (topic involves relationships → graph.md, entities → assets.md, classifications → taxonomies.md, semantic clusters → vectors.md), read those too via the Read tool.

Skip deactivated files per config.

Step 4 — Synthesize briefing

Combine all collected entries into a focused, actionable briefing. This is synthesis, not concatenation — weave information from multiple files into a coherent picture.

Output format:

## Recall: [topic]

[2-3 sentence synthesis — what you need to know to start working on this topic right now]

**Recent**: [latest activity from timeline.md and/or last.md]
**Decisions**: [key decisions with rationale, from decisions.md]
**Lessons**: [relevant lessons, if any, from lessons.md]
**State**: [current progress on this area, from progress.md]
**Next**: [what was planned or pending]

Sources: decisions.md, timeline.md, progress.md [list only files that contributed]

Rules:

• Keep output under 20 lines total
• Omit sections that have nothing relevant (e.g., skip Lessons if no lessons found for this topic)
• Cite source files inline in the Sources: footer
• Synthesize, don't dump — answer "catch me up on X" not "here are all mentions of X"
• If nothing found across any file: No memory found for "[topic]". Try pmm:query [topic] for a deeper search with filters and traversal.

---

Agent Dispatch (eager mode / bootstrap not wired)

Used only when Mode: eager OR bootstrap_wired: false.

Dispatch a general-purpose agent using the Readonly Agent Model from memory/config.md (default: haiku). Replace <project-root> with the actual project root path and <topic> with $ARGUMENTS.

Recall memory for a topic briefing. This is a READ-ONLY task — do not edit any files.

Project root: <project-root> Topic: <topic> (empty = session resume)

If topic is empty — Session Resume

Read <project-root>/memory/last.md and <project-root>/memory/progress.md.

Return:

## Session Resume

**Last**: [1-sentence from last.md]
**State**: [current state from progress.md]
**Next**: [top 2-3 items]

If topic is provided — Focused Recall

1. Read <project-root>/memory/config.md to get active files
2. Read all active Tier 1 .md files in <project-root>/memory/
3. For Tier 2 files (graph.md, vectors.md, taxonomies.md, assets.md): read only if the topic likely involves relationships, entities, or classifications
4. Search all read files for entries related to the topic (case-insensitive)
5. Synthesize a focused briefing:

## Recall: [topic]

[2-3 sentence synthesis]

**Recent**: [from timeline.md, last.md]
**Decisions**: [from decisions.md]
**Lessons**: [from lessons.md, if any]
**State**: [from progress.md]
**Next**: [what was planned]

Sources: [files that contributed]

Keep output under 20 lines for topic recall, under 10 for resume. Omit sections with no relevant entries. If nothing found: No memory found for "[topic]". Try pmm:query [topic] for a deeper search with filters and traversal.

Return the formatted output as a string. Do not write to any file.

Output the agent's return value verbatim.

---

Notes

• Differentiation from pmm:query: query is search (find specific things, filters, attribution, dump mode). recall is briefing (synthesize working context for a topic, actionable output). query answers "what did we decide about X?" — recall answers "catch me up on X so I can start working."
• Context-first path is the default when session_start: lazy and bootstrap_wired: true. This eliminates agent dispatch for in-window recall.
• Agent dispatch is the fallback for eager mode or unwired sessions — file reads happen inside the agent.
• Model selection follows Readonly Agent Model in memory/config.md (default: haiku). No reasoning required for read-only synthesis.
• For the full memory file reference, see references/README.md.