Recall

/recall — Memory Layer active recall

> Active consumption of the memory layer, filling the blind spots of hook-triggered passive recall.

When to use

When you proactively want past context (not a hook firing automatically):

• Stuck while writing a plan/spec — wondering "how did we decide similar cases before"
• Design choice — "X or Y" and want to see the historical rationale
• Cross-module reuse — "which module solved a similar problem before"
• Methodology query — "what is our standard here"
• Historical lesson query — "have we hit this pitfall before"

Not for:

• Literal in-file search → grep
• Function call chains → /graph what-uses / refs
• Third-party library docs → context7
• Git commit history → git log

Input

/recall 
/recall  --k 5            # default k=5
/recall  --kind decision  # filter by source kind
/recall  --path docs/plan # path filter

`` can be:

• A natural-language query: auth cross-tenant isolation
• Keywords: report pipeline release gate
• Filename + keywords: dashboard-overview.ts feature flag

Workflow

Step 1 — Pick K and filter

K defaults to 5 (a fuller view on active recall).

If `` involves:

• decision/proposal → add --kind decision
• standard/spec → add --kind standards
• failure/pitfall → add --kind error,journal
• file history → add --path

Step 2 — Call the retrieve CLI

npx tsx .claude/memory/scripts/retrieve.ts "" --k 5 --format briefing

With a path filter (decisions generally live under docs/plan/ or docs/spec/):

npx tsx .claude/memory/scripts/retrieve.ts "" --k 5 --format briefing --path-filter "docs/(plan|spec)/"

With an arbitrary path filter:

npx tsx .claude/memory/scripts/retrieve.ts "" --k 5 --format briefing --path-filter ""

Step 3 — Present + self-assess

Output the retrieve briefing (top-5 source path + breadcrumb + short preview).

Then self-assess relevance:

• which one directly answers the topic
• which one is only tangentially related
• which one is irrelevant

Output format

### 💾 Relevant Memory (top-5)

1. **[decision]** PLAN-dashboard-overview.md — dashboard overview pipeline...
   `docs/plan/PLAN-dashboard-overview.md` · 0d old

2. **[standards]** ENGINEERING-STANDARD.md — engineering invariants + state machine...
   `docs/standards/ENGINEERING-STANDARD.md` · 7d old

...

Domain-first self-check

The memory layer is a meta-layer, not a domain. /recall usage limits:

• ✅ Before/during plan/spec writing: high value, prevents reinventing the wheel
• ✅ At decision moments: historical decision rationale
• ⚠️ Don't use it for exploratory idle browsing — a slow retrieve doesn't beat grep
• ❌ Don't use it as a retrospective substitute — a retro reads git, not memory chunks

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: AbyssCN
• Source: AbyssCN/oh-my-dag
• License: MIT

Install and usage instructions live in the source repository linked above.