Persistent JSONL knowledge store for cross-session learnings. Saves entries via /dog:learn, searches and lists via /dog:learnings. Use when you want to capture a specific insight, decision, or technique from the current session for future recall.

/dog:learnings — Persistent Cross-Session Knowledge Store

A searchable JSONL database of learnings extracted from sessions. Works alongside the existing dog:learner skill (which saves reusable skill files). This skill stores granular insights — specific techniques, decisions, and patterns that don't warrant a full skill file.

When to Use

When /dog:learn extracted a broad pattern, but you also want to save a specific detail
When you learned something session-specific (e.g., "this repo uses X convention for Y")
When searching for something you remember learning but can't find in skills

Storage

File: ~/.dog/projects/{slug}/learnings.jsonl
Slug: git remote get-url origin | slugify (or $(basename $PWD) if no remote)

JSONL line schema:

{
  "ts": "2026-04-15T12:00:00Z",
  "tags": ["bun", "testing"],
  "title": "Use Bun built-in test runner for TS projects",
  "body": "bun test runs without vitest config. Use `bun test --watch` for TDD.",
  "files": ["package.json"],
  "commit_sha": "abc1234"
}

API

Save a learning (via /dog:learn)

/dog:learn has been extended to also append to the JSONL store after saving the skill file.

The JSONL entry is written with:

title: first line of the learning description
body: full description
tags: auto-extracted keywords (language names, tool names, concept keywords)
files: files mentioned or recently edited
commit_sha: current HEAD

List recent learnings

/dog:learnings               # last 20 entries
/dog:learnings bun           # search by keyword
/dog:learnings --tag testing # filter by tag
/dog:learnings --since 7d    # entries from last 7 days
/dog:learnings --files pkg   # entries mentioning files matching 'pkg'

Search implementation

SLUG=$(git remote get-url origin 2>/dev/null | sed 's|[/:.]|-|g' || basename $PWD)
FILE=~/.dog/projects/${SLUG}/learnings.jsonl

# Keyword search
grep -i "$QUERY" "$FILE" | tail -20 | while IFS= read -r line; do
  echo "$line" | jq -r '"[\(.ts[:10])] \(.title)"'
done

# Tag filter
jq -r 'select(.tags[] | contains("'"$TAG"'")) | "[\(.ts[:10])] \(.title)"' "$FILE"

# Since filter (approximate — compare date prefix)
jq -r 'select(.ts >= "'"$(date -d '-7 days' +%Y-%m-%d 2>/dev/null || date -v-7d +%Y-%m-%d)"'") | "[\(.ts[:10])] \(.title)"' "$FILE"

Process

Saving via /dog:learn

When /dog:learn is invoked, after saving the skill file:

Extract metadata: title, tags (keywords from description), files touched, current commit SHA
Build JSONL entry
Ensure ~/.dog/projects/{slug}/ directory exists
Append entry to learnings.jsonl
Report: Saved to learnings.jsonl: "{title}"

Listing/searching via /dog:learnings

Resolve slug and locate ~/.dog/projects/{slug}/learnings.jsonl
If file doesn't exist: print No learnings found for this project.
Apply filters (keyword, --tag, --since, --files)

Display results as numbered list:

1. [2026-04-15] Use Bun built-in test runner for TS projects
   Tags: bun, testing
   bun test runs without vitest config. Use `bun test --watch` for TDD.

If no results: No learnings matching "{query}".

Verification

SLUG=$(git remote get-url origin 2>/dev/null | sed 's|[/:.]|-|g' || basename $PWD)
tail -1 ~/.dog/projects/${SLUG}/learnings.jsonl | jq .title

Should print the title of the last saved learning.

Skill: dog:learner (saves full skill files; this skill saves granular insights)
Command: /dog:learn (entry point for saving), /dog:learnings (entry point for querying)
Agent: pattern-extractor

learnings