/journal

Write a structured working-memory entry for the current Claude session. Entries land at workspace/threads/journal/<YYYY-MM-DD>/<NNN>-<slug>.md and the day's INDEX.md is auto-updated.

Spec: core/knowledge/public/hq-core/journal-spec.md.

Usage

Form	Effect
/journal "<title>"	Write a new entry inferring body from recent context
/journal --list	Print today's INDEX
/journal --read <NNN>	Print a specific entry

Behavior

Writing an entry

1. Synthesize the body from the current conversation context. The entry captures what's useful to recover if the prefix gets compacted away — not everything that happened.
2. Use this template (50–200 lines, ≤5KB hard):

## Goal
<one sentence: what this slice of work is about>

## Findings
<bullets — non-obvious things learned, file paths discovered, behaviors observed>

## Decisions
<numbered — what was chosen, including rejected alternatives if instructive>

## Next
<what the next entry will pick up: story id, file, or open question>

3. Pass the synthesized body via --body-file to core/scripts/session-journal.sh write:

mktemp_body=$(mktemp -t journal-body.XXXXXX.md)
cat > "$mktemp_body" <<'EOF'
## Goal
...

## Findings
...

## Decisions
...

## Next
...
EOF

core/scripts/session-journal.sh write "<title>" \
  --files "<comma-separated relative paths touched, optional>" \
  --body-file "$mktemp_body"

rm -f "$mktemp_body"

4. Confirm the path returned by the helper. Print the entry's relative path back to the user — that's the receipt.

--list

core/scripts/session-journal.sh list

Prints today's INDEX.md. If the user wants a different day, pass --date YYYY-MM-DD through.

--read <NNN>

core/scripts/session-journal.sh read <NNN>

Prints a specific entry. Useful after /handoff continuation when the SessionStart hook surfaced the INDEX and you need a specific entry's detail.

When to write an entry

You don't need to wait for the auto-trigger hook. Write proactively after:

• Finishing a coherent slice of work (one story, one investigation, one cluster of related edits).
• A commit lands.
• A test pass / typecheck pass.
• A user direction change ("now let's do X instead").
• Right before /handoff or /checkpoint.
• When you notice you've made non-obvious findings that wouldn't survive a compaction.

Constraints

• Never include secrets, share-session URLs, or other capabilities in entries (per .claude/policies/hq-share-session-urls-are-capabilities.md and the rest of the credential-isolation rule set).
• Don't write trivial entries ("read a file"). The bar is: would future-me want this back after compaction?
• One entry per coherent slice. If you have three unrelated findings, write three entries.

Related

• core/knowledge/public/hq-core/journal-spec.md — full spec, two-pattern coexistence with task journals
• .claude/hooks/journal-due.sh — PostToolUse auto-trigger
• .claude/hooks/journal-precompact.sh — PreCompact reminder
• .claude/hooks/load-journal-index-on-start.sh — SessionStart auto-load
• core/scripts/session-journal.sh — helper script
• .claude/skills/_shared/journal.sh — task journal (different pattern, doesn't conflict)