name: abcmemory description: User vocabulary abcmemory and abcremember — where Cursor guidance lives and how to persist new instructions. Use when the user says abcmemory or abcremember.
abcmemory / abcremember
One rule (Cursor-only)
abcmemory is only under .cursor/ (plus .cursorrules and .cursorignore). That is where Cursor loads standing agent guidance.
.llm/ is a planning workspace (plans, optional history, human templates, reference context). It is not abcmemory. Skills under .cursor/ may reference .llm/plans/ for workflow, but abcremember writes to .cursor/ unless you explicitly ask to store something under .llm/.
When to use
- User says abcmemory — they mean the repo's Cursor guidance files (skills, rules, prompts, hooks, etc.).
- User says abcremember — create or update abcmemory with what they told you, then apply it going forward.
Where abcmemory lives
Committed guidance for this repo lives only in:
.cursor/skills/**— oneSKILL.mdper directory.cursor/rules/**— Cursor rules (.mdc).cursor/prompts/**— reusable agent-facing prompt snippets.cursor/hooks/**and.cursor/hooks.json— optional Cursor project hooks (when used).cursorrules— root-level rules.cursorignore— path-level ignores for Cursor
Not abcmemory: .llm/plans/, .llm/history/, .llm/templates/, and .llm/context/ are operator workflow and reference material. See .llm/LLM.md.
Do not duplicate the same guidance under .github/, docs, or .llm/history/ unless the user explicitly asks.
Contributor policy: docs/development/llm/DOCS-DEVELOPMENT-LLM.md. See also llm-cursor-source.
Prompts vs .llm/templates/
.cursor/prompts/— abcmemory; standing or reusable agent prompt snippets (kebab-case.mdfiles)..llm/templates/— human copy-paste templates for plans and optional history notes; not agent instruction source.
abcremember workflow
- Understand what the user wants remembered (scope, always-on vs path-specific vs task workflow).
- Search existing abcmemory (skills, rules, prompts,
.cursorrules) for overlapping guidance — use the placement guide below plus grep/read, not guesswork. - Decide:
- Update when the instruction refines, extends, or corrects existing guidance in the same domain.
- Create when the topic is new, cross-cutting enough to warrant its own skill/rule, or scoped globs need a new
.mdc.
- Write minimally; do not duplicate the same rule in two abcmemory locations.
- Report which files were created or updated so the operator can commit.
How to abcremember (placement guide)
Use this table to choose where to create or update abcmemory:
| Kind of instruction | Prefer |
|---|---|
| Always-on repo convention | .cursorrules or new/updated always-applied .cursor/rules/*.mdc |
| Applies to specific paths/file types | Scoped .cursor/rules/*.mdc with globs |
| Domain/task workflow (API, E2E, K8s, etc.) | New or extended .cursor/skills/<topic>/SKILL.md |
| Reusable prompt block for agents | .cursor/prompts/<topic>.md |
| Cursor event automation (shell gates, etc.) | .cursor/hooks.json + .cursor/hooks/* |
| Path ignores for Cursor indexing | .cursorignore |
| Feature plan, session log, or one-off note | .llm/plans/ or .llm/history/ — not abcremember unless you also want standing policy in .cursor/ |
Rules when abremembering
- One source of truth — do not duplicate the same guidance elsewhere unless the user explicitly asks.
- Match existing repo style (frontmatter, skill
name= folder name, blank line after closing---). - Keep changes minimal and focused on what the user asked to remember.
- After abremembering, tell the user which abcmemory files were created or updated so they can commit them.
Build process + docs sync guardrail
When build orchestration changes (package.json build scripts, CI validate sequencing, or workspace
build runner behavior), abcremember requires updating build-order guidance in both:
.cursor/skills/build-order/SKILL.mddocs/development/tooling/DOCS-DEVELOPMENT-TOOLING-BUILD-ORDER.md
Do not merge build-order changes that leave the canonical build-order document stale.