Learn-skills.dev project-memory
Always-on project memory workflow. At the start of every task, read `.agents/LEARNINGS.md`; while working, update it with mistakes, corrections, successful and failed patterns, and project-specific preferences.
install
source · Clone the upstream repo
git clone https://github.com/NeverSight/learn-skills.dev
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/NeverSight/learn-skills.dev "$T" && mkdir -p ~/.claude/skills && cp -r "$T/data/skills-md/abpai/skills/project-memory" ~/.claude/skills/neversight-learn-skills-dev-project-memory-37b602 && rm -rf "$T"
manifest:
data/skills-md/abpai/skills/project-memory/SKILL.mdsource content
Project Memory
Maintain a per-project markdown memory file so future work improves over time.
This skill is always active. No trigger phrase is required.
Session Start
- Read
before doing any task work..agents/LEARNINGS.md - Apply its guidance silently during execution.
- If it does not exist, create it with this template:
# LEARNINGS ## Corrections | Date | Source | What Went Wrong | What To Do Instead | | ---- | ------ | --------------- | ------------------ | ## User Preferences - (accumulate here as you learn them) ## Patterns That Work - (approaches that succeeded) ## Patterns That Don't Work - (approaches that failed and why) ## Domain Notes - (project/domain context that matters)
Adapt sections to fit the repo, but keep entries easy to scan.
Continuous Updates
Update
.agents/LEARNINGS.md continuously while working, not only at session boundaries.
Add an entry immediately when:
- You identify the cause of an error.
- The user corrects your behavior or preferred approach.
- You catch your own mistake.
- An attempted approach fails.
- A non-obvious approach works reliably.
- You re-check project memory before a risky or previously error-prone step.
What To Log
Log anything that should change behavior in future sessions:
- Self mistakes: wrong assumptions, misread code, failed commands, incorrect fixes.
- User corrections: what was requested instead.
- Tool or environment surprises.
- User preferences: style, workflow, structure.
- Reliable patterns: what worked and when.
Write entries that are specific and actionable.
Bad:
Made an error
Good: Assumed API returned a list; it returns a paginated object with .items
Maintenance
Every 5 to 10 sessions, or when file length exceeds about 150 lines:
- Merge redundant entries.
- Promote repeated corrections into
.User Preferences - Remove notes already captured by stable rules.
- Archive outdated or resolved items.
- Keep the file under 200 high-signal lines.
Examples
Self-caught mistake:
| 2026-02-06 | self | Passed (name, id) to createUser but signature is (id, name) | Check function signatures before calling; this repo does not use conventional arg ordering |
User correction:
| 2026-02-06 | user | Used relative imports | Use absolute imports from `src/` in this repo |