Skill v1.0.0
currentAutomated scan100/100version: "1.0.0" name: multica-platform-jenkins description: Platform-layer skill (placeholder shell): read/write capability for CI/CD systems (Jenkins-class) — trigger parameterized builds, poll status, fetch console logs. Parameters are auto-discovered; never hardcode. Credentials injected via runtime env. Concrete CI URL / Job list in config.yaml, never in role prompts. metadata: layer: platform replaces: any CI/CD system (Jenkins / GitLab CI / GitHub Actions / self-hosted pipeline / etc.) runtime: python: ">=3.10"
Platform · Jenkins (placeholder shell)
This is a platform-layer placeholder shell. The public multica-best-practices binds to no specific company's internal URLs or credentials.To onboard your own CI/CD system, only edit this skill'sconfig.yamlandscripts/; all upstream orchestration skills stay untouched.
Purpose
CI/CD system read + write: trigger Jobs, poll builds, fetch consoleText, return build URLs.
Parameter auto-discovery: connect to the CI system API to read each Job's required parameters; Agents must not hardcode parameter names.
Default CI URL configured in config.yaml → cicd.base_url (placeholder: http://<your-cicd-host>).
Agent standard flow (required)
1. Resolve service (jobs-catalog / issue_service_map) + **deploy branch** (from the Issue source + affected ends, not feature; for linked Issues use the branch in the external system)2. discover (required) — default copy from lastSuccessfulBuild, only override deploy branch3. ready=false → fill missing params or BLOCKED ask human4. trigger: multica-artifact-cicd-sync → trigger script
Parameter priority: --param > branch hint (branchName etc.) > last SUCCESS build params > CI default
Files (suggested structure)
multica-platform-jenkins/├── SKILL.md├── config.yaml├── jobs-catalog.yaml # per-env Job list (service → Job name)├── .env.example└── scripts/├── trigger_env.py # trigger by env + service├── list_jobs.py # list / query Jobs├── cicd_cli.py # low-level API└── lib/
Install deps (once)
pip install -r scripts/requirements.txt
List Jobs
python scripts/list_jobs.py --env devpython scripts/list_jobs.py --env sit --service <service>
Discover parameters (live from CI, per Job)
python scripts/cicd_cli.py discover-params --env sit --service <service> --branch release/<ISSUE>-<slug> --json
Trigger build (auto-params by default)
python scripts/trigger_env.py --env sit --service <service> --branch release/<ISSUE>-<slug> --jsonpython scripts/trigger_env.py --env sit --service <service> --branch release/<ISSUE>-<slug> --param deployVersion=1.0.0_795 --json
Credentials
| Variable | Description | |
|---|---|---|
CICD_USER / CICD_PASSWORD | CI account (runtime env, never in role prompts) | |
CICD_TOKEN | optional token |
Success criteria
- script exit code
0 - every build
result=SUCCESS - return build URL list
Relationship to orchestration skill
| Orchestration skill | Calls | |
|---|---|---|
multica-artifact-cicd-sync | trigger_cicd.py → internally calls this skill's Python scripts |
Why it works
Parameter discovery: different projects have different parameter names, so the Agent discovers then triggers, avoiding hardcoding branchName. Job names live in jobs-catalog.yaml. A swappable platform layer is the core of "copy-paste-run".