
Query and download in-VM Taskcluster worker logs from SolarWinds Observability (formerly Papertrail) using paperctl v2.0. Use when investigating what a worker process or its OS reported on-host. DO NOT USE FOR provisioning failures where no worker started (use tc-logview) or Azure-side VM lifecycle events (use splunk).
Use when validating, auditing, grading, or checking an agent skill, SKILL.md, validator report, eval coverage, or production-readiness claim against agentskills.io and repo conventions. Checks spec, style, links, references, and eval evidence. DO NOT USE FOR creating or iterating skills; use skill-creator.
Use when performing comprehensive failure triage for sheriffs and image maintainers. Automatically determines if a failure is caused by code changes, image changes, or is a known intermittent. Combines data from Taskcluster, Treeherder, and worker image analysis. Triggers on "triage", "sheriff", "why did this fail", "is this an image regression", "failure analysis".
Generate a daily work log by scanning Claude Code (~/.claude/projects/) and Codex (~/.codex/sessions/) session JSONL files and compiling them into ~/moz_artifacts/daily-log-YYYY-MM-DD.md. Use when summarizing a day's sessions or reviewing what the user worked on.
Search, create, modify, comment, link, or transition issues in Mozilla JIRA (mozilla-hub.atlassian.net) via the extract_jira.py script. Markdown → ADF conversion is built in. Prefer this skill over the Atlassian MCP for JIRA work — it handles RELOPS defaults, sprint queries, comment edit/append, and bulk modify with dry-run.
Query a local SQLite database of Windows 11 24H2 and 25H2 cumulative- update file information — DLLs, drivers, and system files across all KB releases. Use when investigating which build introduced a file version, comparing files between patches, tracking a file's version history, or debugging Windows update regressions in CI worker images.
Query Mozilla Splunk Cloud for Azure activity logs (VM lifecycle, NIC/disk writes, deletes, OSProvisioningTimedOut) on Taskcluster worker pools — driving Splunk Web through `browser-harness` against an authenticated Chrome tab because the REST API is disabled at Mozilla. Use for VM-side provisioning failures and worker lifespan on `index=azure_audit`. DO NOT USE FOR in-VM logs (use papertrail) or worker-manager service decisions (use tc-logview).
Run Firefox `mach try` with pre-configured worker pool overrides for alpha-image testing on Windows and Linux pools (win11-24h2, win11-25h2, win11-arm64, win11-amd, win10-2009, b-win2022, win11-source). Use when pushing tests to alpha pools to validate worker images before they ship to production.
Query Mozilla's Redash (sql.telemetry.mozilla.org) for telemetry from BigQuery via saved query IDs or ad-hoc SQL. Use when the task references a saved query, needs FXCI worker-pool queue-time data, or wants results that can be shared and visualized. DO NOT USE FOR raw bq CLI work without a saved query (use bigquery).
Poll Mozilla's public Lando API to check the status of a landing job (submitted, in_progress, landed, failed) and surface the landed commit hash or failure reason. Use after submitting a try push or commit through Lando to verify whether it landed.
Use when searching, viewing, creating, updating, commenting on, or attaching files to Mozilla Bugzilla (bugzilla.mozilla.org) tickets via the bz.py CLI. Use whenever the task involves a Bugzilla bug — filing a regression, triaging a crash, needinfo'ing a reviewer, or following up on assigned bugs.
Query the Taskcluster task graph to list tasks assigned to specific worker types — for worker-pool migrations, audits of which tasks run on a pool, or building targeted `mach try fuzzy` queries. DO NOT USE FOR live task status, retriggers, or task logs (use taskcluster).
Run ad-hoc SQL against Mozilla telemetry tables using the bq CLI. Use when the task needs raw BigQuery access — dry-run cost checks, custom SQL, pulling DAU/MAU/Windows-distribution/Glean-metric data into a script. DO NOT USE FOR saved/shared queries with visualizations (use redash) or finding which probe to query (use mozdata:probe-discovery).
Interact with Mozilla Taskcluster CI: status, logs, artifacts, retriggers, in-tree actions (confirm-failures, backfill, retrigger-multiple), task groups, and worker-pool ops via the native taskcluster CLI plus tc.py. DO NOT USE FOR finding which tasks run on a worker pool (use task-discovery) or worker-manager provisioning logs (use tc-logview, see references/tc-logview.md).
Trigger GitHub Actions workflows in mozilla-platform-ops/worker-images to build Firefox CI Windows worker images (FXCI Azure trusted and untrusted, win10-2009 / win11-24h2 / win11-a64 / win2022). Use to start an image build or check the status of a triggered run. DO NOT USE FOR debugging an existing image's failures (use worker-image-investigation).
House style for authoring and maintaining Mozilla CI agent-skills in this repo. Use when creating, editing, or reviewing a SKILL.md, the description frontmatter, references/, or scripts/ — covers description style, gotchas convention, scope-vs tables for log tools, hard-coded paths, and mozdata: handoffs.
Use when querying Mozilla Treeherder for CI job results, failure analysis, test history, similar-job comparison, cross-branch sheriff triage, job classification, and bug suggestions via treeherder-cli plus the Treeherder REST API for endpoints the CLI doesn't cover.
Investigate Taskcluster task failures caused by worker images — extract image versions, compare passing vs failing tasks, find pass/fail cliffs across branches, debug Azure VMs via `az` CLI, and spin up throwaway debug VMs from the same image a pool uses. Use when a CI failure looks image-related. DO NOT USE FOR triggering a new image build (use worker-image-build).
Download Azure Cost Management exports and query local Parquet/CSV in DuckDB. Use when refreshing local Azure cost caches or writing DuckDB SQL over exports. DO NOT USE FOR live Cost Management API diagnosis; use azure-cost-analysis.
Use when tracing Taskcluster Azure VM startup from worker-manager request through in-VM boot scripts to generic-worker `workerReady` with tc-logview, paperctl, Splunk Web, and Yardstick Prometheus. Applies to Windows worker provisioning latency. DO NOT USE FOR task failure triage (use worker-image-investigation).
Prepare one-on-one/status bullets from ~/moz_artifacts using qmd and copy a topic-organized HTML/RTF list with embedded links to the macOS clipboard. Use when summarizing recent Mozilla work for a manager, 1:1, or status update. DO NOT USE FOR generating raw daily logs; use daily-log.
Use when creating performance self-reviews from local notes, prior reviews, review prompts, and verified evidence. Helps draft H1/H2, annual, and promotion self evaluations, example answers, and rich review-form paste output. Do not use for routine status or 1:1 summaries; use one-on-one.