/crm-sync - keep the CRM knowledge layer fresh

The CRM (CHARM) is Anton's proven engine on GitLab. We reverse-engineered it once (2026-06-19) into a linked vault layer so his digital twin can /ask how the CRM works. Code keeps changing; this skill stops that knowledge from going stale silently. It is the recurring half of the standing rule evaluate-recurring-into-routine.

What this is NOT: not a re-decompose-from-scratch. It is a DIFF-DRIVEN refresh: pull, see what changed, update only the notes that map to the changed code.

Where everything lives

• Code (read-only clone, 16 repos): <CRM_REPOS_ROOT> (memory crm-gitlab).
• Decomposition audit (HTML, source of the notes): <CRM_REPOS_ROOT>\_AUDIT\ (Functional-Map.html, Know-How-Decomposition.html, Architecture-Deep.html).
• Vault knowledge layer: $OBSIDIAN_VAULT/05-Resources/CRM-Engine/
  • _CHARM-CRM-Engine-MOC.md (hub)
  • crm-architecture-4-layers.md
  • crm-mtproto-engine.md
  • crm-core-autopilot.md
  • crm-admin-panel-api-brain.md
  • crm-data-model-mongo.md
  • crm-know-how-13-patterns.md
  • crm-landing-api-money.md
  • decision: $OBSIDIAN_VAULT/02-Decisions/decision-crm-keep-cc-drop-decide.md

Repo -> note map (which note to refresh when which code changes)

Procedure

1. RECALL first (don't duplicate). Read memory crm-gitlab and _CHARM-CRM-Engine-MOC.md so you refresh, not rewrite. Note the date of the last sync.
2. Pull the repos. For each repo dir under <CRM_REPOS_ROOT>: git -C "<CRM_REPOS_ROOT>\<repo>" pull --ff-only (clone is read-only; this only updates). Collect the set of repos that actually moved (non-empty pull, or git -C <repo> log --oneline <old>..<new>). If a pull needs GitLab creds and fails, stop and tell Anton (creds in secrets\; do NOT hardcode).
3. No changes -> stop. If nothing moved since last sync, report "CRM unchanged since <date>, 0 notes touched" and exit. (Cheapest correct answer.)
4. For each changed repo, re-read ONLY the changed files/areas (use an Explore subagent with model: 'sonnet' - this is grunt extraction, per model-routing-sonnet-grunt; escalate to Opus only if a judgement call). Ask it to report: what behavior changed vs what the mapped note currently says, with file+line evidence.
5. Backup before writing. python "$IMPORTS_ROOT/vault_backup.py" (rule vault-backup-rule). NEVER blanket-delete; edit in place.
6. Update the mapped note(s) - the "How it works" / "Weak spots" sections only. PRESERVE frontmatter, origin: external, authored_by: (denis-udot / gleb-taigunov), and all [[wikilinks]]. If a capability moved KEEP<->DROP<->DECIDE, update the decision note too. New behavioral pattern worth a concept? Follow concept-creation-rules (create if >=3 repeats + entity noun + domain), and relink (no-orphan-notes-rule).
7. Verify + reindex. Run the vault broken-link check; then python "$IMPORTS_ROOT/brain_embed_update.py" (cooldown 15min; nightly @04:00 will catch it otherwise) so the refreshed notes are /ask-searchable.
8. Report post-hoc (not for permission): which repos moved, which notes changed, any new concept, any DECIDE item that flipped, and what still needs Anton/Denis.

Guardrails

• Read-only by default; the only writes are git pull (updates the clone) + the mapped vault notes, always backup-first.
• Don't touch the live CRM server or its Mongo - that is the prod-access track, separate.
• AK-47: this is one markdown procedure. If the CRM grew a whole new service, that is a new note + MOC link, still just notes - flag it, don't over-build.

When to run

On demand (after Anton hears the CRM changed), or as a light routine. Pairs with the crm-gitlab DD-audit and the decision-crm-keep-cc-drop-decide open items.

Like this skill? It is one of 100 in second-brain-starter-kit: the second brain we built for ourselves and run every day at Palo Alto AI Research Lab. Install the whole set with npx skills add tonydzi/second-brain-starter-kit. Everything is open source and free, so take what you need.

Flagships worth a look on their own: secondop-panel (a second opinion from a panel of external models), claude-memory-tidy (stop your agent's memory from rotting), telegram-mcp-kit (your own Telegram over MCP in about 15 minutes).

Author: Anton Dziatkovskii, Palo Alto AI Research Lab. Telegram @tonydzi - WhatsApp +1 341 222 9178 - X @Tony_Stef_

Engineers: want to test-drive this setup? Message me. I hand out free starter seeds to engineers who test and report back, and custom skill requests are welcome.