1c-project-init

Source workspace: C:\Users\YOUR_USERNAME\workspace\ai\1c-AI-workspace

Mode detection

• .claude/skills/ exists in target → enrich (sync missing/outdated skills & docs)
• No .claude/skills/ → new (full init)

CRITICAL: Use skills, not manual 1C commands

NEVER run 1cv8 DESIGNER manually when a skill exists for the operation.

| Task | WRONG | RIGHT | |------|-------|-------| | Dump XML | ssh ... docker exec ... 1cv8 DESIGNER /DumpConfigToFiles | /db-dump-xml skill | | Load XML | docker exec ... 1cv8 DESIGNER /LoadConfigFromFiles | /db-load-xml skill | | Build CF | manual 1cv8 | /db-dump-cf, /cf-init skills |

The /db-dump-xml skill reads .v8-project.json automatically and connects to the remote server directly — no need to run DESIGNER inside the container. License PROF on the local machine is sufficient for all DESIGNER operations including DumpConfigToFiles and extension dumps.

Steps

1. Determine target path

• If argument provided → use it
• Otherwise → current working directory (pwd)

2. Run init script

powershell.exe -NoProfile -File "C:\Users\YOUR_USERNAME\workspace\ai\1c-AI-workspace\.claude\skills\1c-project-init\scripts\init.ps1" -TargetPath "<target>" -Mode <new|enrich>

Script output lists what was copied/updated.

3. If mode = new — collect project info interactively

Ask ALL questions upfront in a single AskUserQuestion call (multiple questions at once).

Question order matters — project type and extension names come FIRST:

1. Project type (options): configuration / extension / configuration + extension / external-processor / other
2. Extension name(s) — ask only if type includes extension. If multiple, comma-separated (e.g. МоёРасширение, ДругоеРасширение)
3. 1C base name (e.g. Forte, Buh)
4. Platform version: 8.3.24 / 8.3.25 / 8.3.27
5. Web publication name (e.g. forte, buh) — for HTTP API and playwright
6. Admin login and password — for .v8-project.json

Note: questions 1+2 can be asked together in first call; 3-6 in second call if user answered that extensions are involved. Or ask all in one call if possible.

Platform → server mapping (all on CT107 / YOUR_EDT_SERVER):

• 8.3.24 → server YOUR_EDT_SERVER:1641, container onec-server-24, web port 8081
• 8.3.25 → server YOUR_EDT_SERVER:1541, container onec-server-25, web port 8080
• 8.3.27 → server YOUR_EDT_SERVER:1741, container onec-server-27, web port 8082

4. Generate CLAUDE.md

Read template: C:\Users\YOUR_USERNAME\workspace\ai\1c-AI-workspace\.claude\skills\1c-project-init\templates\CLAUDE.md.template Fill all placeholders:

• {{PROJECT_NAME}} — project name
• {{PROJECT_DESCRIPTION}} — brief description (include project type + extensions if applicable)
• {{V8_VERSION}} — platform version (e.g. 8.3.25)
• {{SERVER}} — YOUR_EDT_SERVER
• {{PORT}} — from platform map
• {{SERVER_SUFFIX}} — 24 / 25 / 27
• {{WEB_PORT}} — from platform map
• {{BASE_NAME}} — 1C base name
• {{PUBLICATION}} — web publication name
• {{EXTENSION_LINE}} — - Extensions: <names> if extensions present, else omit line
• {{DEV_ZONE}} — src/ / ext/<ExtName>/ / src/ + ext/<ExtName>/
• {{DEV_ZONE_NOTE}} — All modifications go into the extension, not the base configuration. if extension-only or mixed; empty if configuration-only
• {{STRUCTURE_LINES}} — include src/ line if has config, ext/ line if has extensions:
  • config: src/ — configuration source
  • extension: ext/<ExtName>/ — extension source (primary development zone)
  • both: src/ — base configuration source (read-only reference)\next/<ExtName>/ — extension source (PRIMARY development zone)

Write to <target>/CLAUDE.md.

5. Generate .mcp.json

Read template: C:\Users\YOUR_USERNAME\workspace\ai\1c-AI-workspace\.claude\skills\1c-project-init\templates\mcp.json.template Fill all placeholders:

• {{PROJECT_NAME}} — project directory name (for bsl-lsp container)
• {{PUBLICATION}} — web publication name (for ai-debug URL)
• {{WEB_PORT}} — from platform map (8081/8080/8082)
• {{ADMIN_USER}} — admin login
• {{ADMIN_PWD}} — admin password
• {{CODEMETADATA_URL}} — codemetadata MCP URL. Format: http://YOUR_MCP_SERVER:<PORT>/mcp. Port from registry: next free in 7530+ range. If codemetadata not yet deployed, use placeholder http://YOUR_MCP_SERVER:{{CODEMETADATA_PORT}}/mcp and note in report.

Write to <target>/.mcp.json.

6. Generate .v8-project.json if not exists

Read template: C:\Users\YOUR_USERNAME\workspace\ai\1c-AI-workspace\.claude\skills\1c-project-init\templates\v8-project.json.template Fill all placeholders:

• {{V8PATH}} — C:\\Program Files\\1cv8\\<V8_VERSION>\\bin matching the server platform. NEVER leave empty — local default may be a different version.
• {{PORT}} — from platform map (1641/1541/1741)
• {{BASE_NAME}} — 1C base name
• {{ADMIN_USER}} — admin login
• {{ADMIN_PWD}} — admin password
• {{WEB_PORT}} — from platform map (8081/8080/8082)
• {{PUBLICATION}} — web publication name

Write to <target>/.v8-project.json.

7. Create openspec structure if mode = new

openspec/
  project.md        ← project context for AI
  changes/          ← active proposals
  specs/            ← feature specs
  archive/          ← done
dist/               ← built artifacts (CFE/CF/EPF) — tracked in git for distribution

8. Deploy infobase on CT107 (mode = new)

Ask: "Развернуть базу на CT107?"

If yes:

Step 8.1 — Copy .dt file to CT107 (/mnt/data/data-24/ maps to /var/lib/onec/ inside container):

scp "<dt_file_path>" root@YOUR_EDT_SERVER:/mnt/data/data-<SERVER_SUFFIX>/<BASE_NAME>.dt

Step 8.2 — Ensure Xvfb is running in the target container (required for 1cv8 DESIGNER):

ssh root@YOUR_EDT_SERVER "docker exec onec-server-<SERVER_SUFFIX> pgrep Xvfb || docker exec -d onec-server-<SERVER_SUFFIX> Xvfb :99 -screen 0 1024x768x16"

deploy-infobase.sh uses DISPLAY=:99 but does NOT auto-start Xvfb. Without this step, deploy fails with "Unable to initialize GTK+".

Step 8.2b — Run deploy script on CT107:

ssh root@YOUR_EDT_SERVER "/opt/1c-dev/deploy-infobase.sh --name <BASE_NAME> --dt /mnt/data/data-<SERVER_SUFFIX>/<BASE_NAME>.dt --server <SERVER_SUFFIX> --admin-user <user> --admin-pwd <pwd>"

Script /opt/1c-dev/deploy-infobase.sh — universal script (supports 24/25/27):

• --server 24|25|27 — selects container (onec-server-XX), ports, web container automatically
• Auto-detects platform version (ls /opt/1cv8/x86_64/ inside container)
• Auto-detects cluster ID via rac cluster list
• DT path mapping: /mnt/data/data-<SERVER>/X.dt → /var/lib/onec/X.dt inside container

Does 4 steps automatically:

1. Check if infobase already exists (rac infobase summary list); if not — create via rac infobase create + PostgreSQL DB (--create-database). Fails with error if UUID can't be obtained.
2. 1cv8 DESIGNER /RestoreIB — restores .dt. Verifies "успешно завершена" in log, exits on failure.
3. onec-webinst.sh publish — publishes web in onec-web-<SERVER_SUFFIX>
4. rac infobase update --license-distribution=allow

Script behaviour:

• DB password read from /opt/1c-dev/secrets/pg_password (not hardcoded)
• Lock file /tmp/deploy-<NAME>.lock prevents concurrent runs
• Log written to /tmp/deploy-<NAME>-<timestamp>.log (timestamped, not overwritten)
• set -euo pipefail — stops on any error

Known cluster IDs: 8.3.24=eecfed03-e569-4bc7-a729-b8518e4b6859, 8.3.27=0c2ac9c1-8017-4ac8-8b93-35f1409a3ecb. Result URL: http://YOUR_EDT_SERVER:<WEB_PORT>/<BASE_NAME>/

Step 8.3 — Verify infobase availability:

# 1. Infobase registered in cluster
ssh root@YOUR_EDT_SERVER "docker exec onec-server-24 /opt/1cv8/x86_64/<ONEC_VERSION>/rac localhost:1645 infobase summary --cluster=<CLUSTER_ID> | grep -i '<BASE_NAME>'"

# 2. Web publication responds
curl -s -o /dev/null -w "%{http_code}" "http://YOUR_EDT_SERVER:<WEB_PORT>/<BASE_NAME>/"
# Expected: 200 or 302

# 3. HTTP API available (if 1c-ai-debug configured)
curl -s "http://YOUR_EDT_SERVER:<WEB_PORT>/<BASE_NAME>/hs/ai/ping" -u "<user>:<pwd>"
# Expected: {"status":"ok"} or any 200 response

Only proceed to next steps if web check returns 200/302. If 503/404 — check deploy log on CT107: docker exec onec-server-24 cat /var/lib/onec/restore_<BASE_NAME>.log

8.3.1. Install AI_Debug extension (MANDATORY for all 1C projects)

AI_Debug extension provides: HTTP API for MCP (1c-ai-debug), unit test framework (ЮТТесты/ЮТУтверждения), and hosts all *_Test modules. Without it, 1c-ai-debug MCP and /1c-test-runner won't work.

Source: C:\Users\YOUR_USERNAME\workspace\ai\1c-ai-debug-extension\dist\AI_Debug.cfe (pre-built, kept in git).

# Load extension into database
powershell.exe -NoProfile -File .claude/skills/db-load-xml/scripts/db-load-xml.ps1 `
  -V8Path "C:\Program Files\1cv8\<V8_VERSION>\bin" `
  -InfoBaseServer "YOUR_EDT_SERVER:<PORT>" `
  -InfoBaseRef "<BASE_NAME>" `
  -UserName "<ADMIN_USER>" -Password "<ADMIN_PWD>" `
  -ConfigDir "C:\Users\YOUR_USERNAME\workspace\ai\1c-ai-debug-extension\src" `
  -Extension "AI_Debug" -Mode Full

Then update DB to apply extension:

powershell.exe -NoProfile -File .claude/skills/db-update/scripts/db-update.ps1 `
  -V8Path "C:\Program Files\1cv8\<V8_VERSION>\bin" `
  -InfoBaseServer "YOUR_EDT_SERVER:<PORT>" `
  -InfoBaseRef "<BASE_NAME>" `
  -UserName "<ADMIN_USER>" -Password "<ADMIN_PWD>"

Verify: curl -s "http://YOUR_EDT_SERVER:<WEB_PORT>/<BASE_NAME>/hs/ai/ping" -u "<user>:<pwd>" → {"status":"ok"}.

Test modules go HERE — never into the main project configuration. See "Test modules (*_Test)" rule in global CLAUDE.md.

8.3.2. Deploy codemetadata MCP (MANDATORY for all 1C projects)

Follow the codemetadata MCP — auto-deploy section in ~/.claude/1c-development-rules.md.

Summary:

1. Find XML config dump in project (src/cf/, src/src/, or src/cfe/)
2. Copy to CT104: ssh proxmox "pct exec 104 -- mkdir -p /mnt/onec-xml/<PROJECT>/src/cf ..."
3. Add service to C:\Users\YOUR_USERNAME\workspace\ai\1c-enhanced-codemetadata\docker-compose-ct104.yml with next free port (7530+)
4. Deploy: scp compose file → docker compose up -d <project>-codemetadata-enhanced
5. Update {{CODEMETADATA_URL}} in .mcp.json with actual port
6. Verify: curl -s "http://YOUR_MCP_SERVER:<PORT>/health"

Port registry (from 1c-development-rules.md): minimkg=7530, mcparqa24=7540, ai-debug=7550, kaf=7560, next=7570, ...

If source dump not yet available (step 8.4 hasn't run yet), defer codemetadata deploy to AFTER step 8.4 completes.

8.4. Export source from deployed database

Use /db-dump-xml skill — reads .v8-project.json automatically, runs local DESIGNER against remote server. Do NOT run DESIGNER manually inside the container — the skill works fine from the local machine with PROF license.

8.4.1 — Export main configuration (if project type = configuration or both):

powershell.exe -NoProfile -File .claude/skills/db-dump-xml/scripts/db-dump-xml.ps1 `
  -V8Path "C:\Program Files\1cv8\<V8_VERSION>\bin" `
  -InfoBaseServer "YOUR_EDT_SERVER:<PORT>" `
  -InfoBaseRef "<BASE_NAME>" `
  -UserName "<ADMIN_USER>" -Password "<ADMIN_PWD>" `
  -ConfigDir "src" -Mode Full

IMPORTANT: Always pass -V8Path matching the server platform version. The local default 1cv8.exe may be a newer version (e.g. 8.3.27), causing "version mismatch" error if server is 8.3.24 or 8.3.25.

Note: Full config dump can take several minutes (17k+ XML files is normal for ERP/BSP-based configs).

8.4.2 — Export extension(s) (if project type = extension or both):

For each extension name in EXTENSION_NAMES:

powershell.exe -NoProfile -File .claude/skills/db-dump-xml/scripts/db-dump-xml.ps1 `
  -InfoBaseServer "YOUR_EDT_SERVER:<PORT>" `
  -InfoBaseRef "<BASE_NAME>" `
  -UserName "<ADMIN_USER>" -Password "<ADMIN_PWD>" `
  -ConfigDir "ext/<EXTENSION_NAME>" -Extension "<EXTENSION_NAME>" -Mode Full

If project type = extension only — skip step 8.4.1, only export extension(s). If project type = configuration + extension — do both 8.4.1 and 8.4.2.

9. Initialize git repository

# 1. Create .gitignore
cat > .gitignore << 'EOF'
# 1C binaries (intermediate/working files)
*.dt

# Built artifacts in dist/ are tracked (for distribution)
# All other CF/CFE/EPF/ERF outside dist/ are ignored
*.cf
*.cfe
*.epf
*.erf
!dist/*.cf
!dist/*.cfe
!dist/*.epf
!dist/*.erf

# Credentials (contains passwords)
.v8-project.json

# Logs and temp
*.log
tmp/

# OS
.DS_Store
Thumbs.db
EOF

# 2. Create .gitattributes
cat > .gitattributes << 'EOF'
* text=auto eol=lf
src/**/*.xml text eol=lf
src/**/*.mdo text eol=lf
*.ps1 text eol=crlf
*.dt binary
*.cf binary
*.cfe binary
*.epf binary
*.erf binary
EOF

# 3. Init repo
git init
git config user.name "Arman"
git config user.email "arman@localhost"

# 4. Create Gitea repo via API
curl -s -X POST "http://YOUR_SERVER:3000/api/v1/user/repos" \
  -H "Authorization: token YOUR_GITEA_TOKEN" \
  -H "Content-Type: application/json" \
  -d "{\"name\":\"<PROJECT_NAME>\",\"description\":\"<PROJECT_DESCRIPTION>\",\"private\":true}"

# 5. Add remote, commit, push
git remote add gitea http://admin:admin123@YOUR_SERVER:3000/admin/<PROJECT_NAME>.git

# Add dirs based on project type:
# configuration only:   git add .gitignore .gitattributes CLAUDE.md .mcp.json openspec/ .claude/ src/
# extension only:       git add .gitignore .gitattributes CLAUDE.md .mcp.json openspec/ .claude/ ext/ dist/
# configuration + ext:  git add .gitignore .gitattributes CLAUDE.md .mcp.json openspec/ .claude/ src/ ext/ dist/

git commit -m "Initial commit: project scaffold + source"
git push -u gitea master

Note: git push with 17k+ files takes time. Run in background if needed. Note: .v8-project.json is intentionally excluded (contains passwords) — keep it only locally.

10. Configure mcp-bsl-lsp on CT100

The mcp-lsp-<PROJECT_NAME> container was created by the init script but needs:

10.1 — Clone project to CT100 workspace:

ssh root@YOUR_SERVER "git clone http://admin:admin123@YOUR_SERVER:3000/admin/<PROJECT_NAME>.git /opt/workspace/<PROJECT_NAME>"

10.2 — Recreate container with correct WORKSPACE_ROOT:

• If extension project: WORKSPACE_ROOT=/projects/<PROJECT_NAME>/ext/<EXTENSION_NAME>
• If configuration only: WORKSPACE_ROOT=/projects/<PROJECT_NAME>/src
• If both: WORKSPACE_ROOT=/projects/<PROJECT_NAME>/src (LSP indexes base config; ext files accessed via Grep)

ssh root@YOUR_SERVER "
docker stop mcp-lsp-<PROJECT_NAME> && docker rm mcp-lsp-<PROJECT_NAME> && \
docker run -d \
  --name mcp-lsp-<PROJECT_NAME> \
  --restart unless-stopped \
  -v /opt/workspace:/projects:rw \
  -e MCP_LSP_BSL_JAVA_XMX=4g \
  -e MCP_LSP_BSL_JAVA_XMS=1g \
  -e MCP_LSP_LOG_LEVEL=error \
  -e FILE_WATCHER_MODE=polling \
  -e FILE_WATCHER_INTERVAL=30s \
  -e HOST_PROJECTS_ROOT=/opt/workspace \
  -e PROJECTS_ROOT=/projects \
  -e WORKSPACE_ROOT=/projects/<PROJECT_NAME>/<DEV_PATH> \
  mcp-lsp-bridge-bsl:latest
"

10.3 — Set up auto-sync (git pull on CT100 every 5 min):

ssh root@YOUR_SERVER "(crontab -l 2>/dev/null; echo '*/5 * * * * cd /opt/workspace/<PROJECT_NAME> && git pull --quiet origin master 2>/dev/null') | crontab -"

Note: .mcp.json template uses /usr/bin/mcp-lsp-bridge (correct path in container).

11. Report

List what was created/updated. Remind user to configure:

• v8path in .v8-project.json if auto-detect fails
• Open project with claude in target directory (MCP servers load from .mcp.json)

Enrich mode specifics

Compare skill files by content hash — copy only if workspace version is newer or file missing in target. Always overwrite .claude/docs/ (platform specs don't change per-project). Never overwrite: CLAUDE.md, .mcp.json, .v8-project.json, openspec/.