Y Search Codex

Inputs

• Accept a session file name or full path from the user.
• Default session root to ~/.codex/sessions when only a file name is provided.
• Confirm the resolved file exists before running search commands.

Search Strategy

Each event is a single long JSON line. Avoid raw full-line searches that flood output.

1. Count matches first.
2. Pull short snippets around matches.
3. Narrow snippets with a second filter.
4. Open full context only after locating a precise target.

Always use at least one output limiter:

• -c for counts
• -o '.{0,60}pattern.{0,60}' for snippets
• -M 200 or -M 500 to truncate long lines
• | head -20 to cap result count

Core Commands

# 1) Count first
rg -c 'search_term' ~/.codex/sessions/<session-file>.jsonl

# 2) Snippets with local context
rg -o '.{0,60}search_term.{0,60}' ~/.codex/sessions/<session-file>.jsonl | head -20

# 3) Narrow further
rg -o '.{0,60}search_term.{0,60}' ~/.codex/sessions/<session-file>.jsonl | rg 'new_string'

# 4) Full context only when needed
rg '"name":"Edit".*search_term' ~/.codex/sessions/<session-file>.jsonl -M 500

Never run:

rg 'pattern' ~/.codex/sessions/<session-file>.jsonl

Useful Patterns

# Human user inputs
rg -o '.{0,40}"type":"user".{0,80}"userType":"external".{0,40}' <session.jsonl> | head -10

# Tool edits
rg -c '"name":"Edit"' <session.jsonl>
rg -o '.{0,60}"name":"Edit".{0,60}' <session.jsonl> | head -10

# Commands run by shell tools
rg -o '.{0,100}"command":"[^"]*".{0,40}' <session.jsonl> | head -20

# Mentions tied to file edits
rg -o '.{0,60}auth.{0,60}' <session.jsonl> | rg 'file_path'

JSONL Reference

Use these keys when filtering:

• Message roles: "type":"user", "type":"assistant", "type":"tool_result"
• Human messages: "userType":"external"
• Tool blocks: "type":"tool_use"
• Common tool names: "name":"Edit", "name":"Write", "name":"Bash", "name":"Task"
• Useful fields: "timestamp", "agentId", "input"

Workflow

1. Resolve session path from user input (<name>.jsonl or full path).
2. Ask for a target term if none is provided.
3. Run count, then snippets, then narrowed snippets.
4. Run full-context grep only for promising matches.
5. Report concise findings with quoted snippets and exact command(s) used.

Safety

• Keep searches read-only.
• Redact secrets if surfaced in snippets.
• Prefer targeted patterns over broad regex to reduce accidental disclosure.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.