Recall

Recall — Session Memory

Retrieve relevant history, rules, and anti-patterns from past sessions. Direct execution.

Design rationale: This skill executes directly rather than spawning subagents because memory retrieval is a simple command sequence (~200 tokens), not substantial analytical work. Per Lita research: "Simple agents achieve 97% of complex system performance with 15x less code."

When This Applies

Default: Retrieve context before any non-trivial implementation.

Tool Reference

Commands

Critical Rule

Always use --robot or --json. Never run bare cass.

Bare cass launches a TUI that will hang AI agents.

Execution Flow

Execute these steps directly. No subagents needed.

Step 1: Distilled Context (Always Start Here)

cm context "{task_description}" --json

Returns:

• Rules — Distilled patterns from past sessions
• Anti-patterns — What NOT to do (and why)
• Suggested searches — Specific CASS queries for more detail
• Historical context — Related past work

Example:

cm context "implement OAuth login" --json
cm context "add new database migration" --json
cm context "refactor the payment module" --json

Step 2: Review and Extract

From cm context output, extract:

Step 3: Deep Dive (If Needed)

If cm context suggests searches or you need more detail:

Search for specific patterns:

cass search "{pattern}" --robot --fields minimal --limit 5

View a specific session:

cass view /path/to/session.jsonl --json

Expand around a match:

cass expand /path -n {line} -C 3 --json

Step 4: Apply Context

Before implementing:

• Rules noted and will be followed
• Anti-patterns noted and will be avoided
• Prior solutions reviewed for reuse
• Gaps identified for grounding

Search Patterns

Task-based

cm context "implement user authentication" --json
cm context "fix pagination bug" --json
cm context "add API endpoint for X" --json

Pattern-based

cm context "error handling in API routes" --json
cm context "database transaction patterns" --json
cm context "form validation" --json

Problem-based

cm context "timeout errors in background jobs" --json
cm context "memory leak in long-running process" --json

CASS Search Reference

Basic Search

cass search "query" --robot --limit 5

Lean Output (fewer tokens)

cass search "query" --robot --fields minimal --limit 5

With Summary

cass search "query" --robot --fields summary

Token-Budgeted

cass search "query" --robot --max-tokens 2000

Workspace-Specific

cass search "query" --workspace "/path/to/project" --robot

By Time Range

cass search "query" --robot --since 7d

Timeline Commands

# Today's sessions
cass timeline --today --json

# Last week
cass timeline --since 7d --json

# Recent activity
cass timeline --days 7 --json --limit 10

Output Interpretation

Rules (high signal)

{
  "rules": [
    {"pattern": "auth", "rule": "Always use bcrypt cost >= 12"}
  ]
}

Follow these. They're distilled from successful sessions.

Anti-patterns (high signal)

{
  "antipatterns": [
    "Don't use localStorage for tokens - session abc failed"
  ]
}

Avoid these. They caused failures before.

Suggested searches (medium signal)

{
  "suggested_searches": [
    "cass search 'jwt refresh' --robot"
  ]
}

Run these if you need more detail on a specific aspect.

Output Formats

--robot              # Default structured output
--robot-format jsonl # Streaming format
--robot-format compact # Minimal single-line JSON
--fields minimal     # Reduce output size
--fields summary     # Just summaries

Query Tips

Search Refinement

Integration with Workflow

At session start (via /prime)

cm context "{project_name}" --json

After claiming a bead (via /advance)

cm context "{bead_title}" --json

Before implementation

cm context "{what_you're_about_to_build}" --json

When stuck

cm context "{problem_description}" --json
cass search "{error_message}" --robot

Troubleshooting

Search returns nothing

# Rebuild index
cass index --full

# Health check
cass health
cm doctor

Export for reference

cass export /path/session.jsonl --format markdown -o reference.md
cass export /path/session.jsonl --format json

When to Use /recall vs Other Tools

Quick Reference

# Distilled context (always start here)
cm context "task description" --json

# Health check
cm doctor

# Search sessions
cass search "query" --robot --limit 5
cass search "query" --robot --fields minimal

# View session
cass view /path.jsonl --json

# Expand with context
cass expand /path -n 42 -C 3 --json

# Timeline
cass timeline --today --json
cass timeline --since 7d --json

# Index
cass index --full

Anti-Patterns

See Also

• /prime — Session startup (includes recall)
• /advance — Bead workflow (includes recall after claiming)
• /ground — External documentation search