allure-notifications (CLI)

Public npm bin @qa-guru/allure-notifications for line 6.0.*.

npx @qa-guru/[email protected] send --config config.json --dry-run
npx @qa-guru/[email protected] send --config config.json \
  --allure-folder build/reports/allure-report/allureReport/awesome \
  --allure-results-folder build/allure-results \
  --project Multistack \
  --build-url "$BUILD_URL" \
  --live

| Flag | Role | |------|------| | send --config <path> | Load config, collage via @qa-guru/allure-notifications-core | | --dry-run | Render PNG; list messengers that would send; no network (default) | | --mock | Render PNG; record mock deliveries; no network | | --live | Live Telegram sendPhoto (ADR 008); needs env token | | --out <path> | Write PNG buffer to disk | | --allure-folder, --allure-results-folder | Override report/results paths from the consumer cwd | | --project | Override base.project | | --report-url, --dashboard-url, --testops-url, --build-url | Override base.links |

Default without --mock / --live is safe dry-run. Live credentials: TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID, TELEGRAM_TOPIC_ID — see docs/telegram-dogfood.md.

Relative paths stored in config resolve from the config file directory. Relative path overrides resolve from the process cwd. Overrides are applied in memory; send never writes a runtime config or copies credentials into JSON.

Offline suggest (workspace / pre-release)

pnpm build
pnpm exec allure-notifications suggest --results packages/core/test/fixtures/dogfood-results
pnpm exec allure-notifications suggest --results packages/core/test/fixtures/dogfood-results \
  --config config/config.dogfood-telegram-full.json --out suggested.json
pnpm exec allure-notifications suggest --results allure-results --profile default

| Flag | Role | |------|------| | suggest --results <dir> | Required raw results directory; no generated report required for suggestion | | --profile default\|kit | Override automatic profile selection | | --config <path> | Existing config as local summary/history/QG/payload facts; never modified | | --out <path> / --write <path> | Create a new config JSON file instead of stdout; existing files are not overwritten | | --signals | Print the collected compact signals JSON instead of a config (for the builder Suggest popover); --profile sets its profile field; no layout, no LLM call |

The output contains only base (paths, canonical chart and dark mode), validated by the same parseConfig used by send. Messenger blocks, credentials and unknown metadata are not copied from --config. Without an output file, stdout is JSON only; validation errors / warnings go to stderr. There is no PNG rendering, sending, LLM or network I/O. Send modes and send-only overrides are rejected for suggest.

Counters come from a local or configured summary.json (widgets/summary.json also works); without one they are derived from *-result.json statuses. History is discovered as history.jsonl in results, their parent or the report directory, not in an unrelated cwd. Explicit config paths win. Invalid/empty history records are ignored; at least two usable runs are required for trend recommendations.

Neighbouring allureQualityGate.json, sonarQualityGate.json (raw Sonar projectStatus) and testsTable.json are local payload facts. AQG / tests-table widgets are also discovered under widgets/kit-panels/. Sonar widget payloads are not treated as raw projectStatus. Paths in emitted JSON are absolute, so moving --out does not change their resolution. Raw results still need allure generate before send can read its summary. A QG chosen only from qualityGate.rules emits a warning: provide its payload or generate the AQG report widget before sending. problemsDistribution needs environment-labelled history to render its heatmap.

Optional LLM advisor (tier 1, ADR 020): disabled unless both env vars are set. When enabled, compact signals go to an OpenAI-compatible chat/completions endpoint, which answers with catalog panel ids; the reply is validated against PANEL_CATALOG and materialized into the same canonical templates (the LLM never picks coordinates). Timeout 15 s. Any error — HTTP, JSON, unknown id, kit-only under default — falls back to the deterministic scorer with a warning: on stderr. The LLM sees only compact signals, never raw results or messenger secrets.

| Env | Role | |-----|------| | ANB_AI_BASE_URL | OpenAI-compatible base, e.g. http://localhost:11434/v1 | | ANB_AI_MODEL | Model name sent in the request | | ANB_AI_API_KEY | Optional credential — Authorization: Bearer by default; a user:pass value is sent as HTTP Basic (e.g. ollama-box2.qa.guru) | | ANB_AI_TIMEOUT_MS | Optional request timeout in ms (default 60000; remote CPU inference can be slow) |

Template capacities and selection order: config package. Builder UI and automatic delivery are separate increments.

Alternate (Allure 3 plugin): same collage + messengers via allurerc done hook — examples/allurerc.notifications.mjs · packages/plugin/README.md. CLI pin stays primary for consumers.

Workspace (pre-publish / local):

ALLURE_RESULTS_DIR="$(mktemp -d)" pnpm --filter @qa-guru/allure-notifications test
pnpm exec allure-notifications send --config … --dry-run