Menu
Support deep dive analysis of Bugzilla reports within specific Products and Components.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
File a Bugzilla bug directly via the BMO REST API with full power the prefilled enter_bug.cgi form can't do — create the bug AND upload one or more attachments (logs, patches, testcases, screenshots) AND set bug-/attachment-level flags (needinfo?, review?, sec-approval?, approval-mozilla-*?, tracking flags) in one confirmed operation. Also attaches files to, and sets flags on, existing bugs. Use when the user wants to actually submit a bug (not just open a draft form), attach files while filing, or set flags. For a simple file-it-myself-in-the-browser draft with no attachments, prefer the form-based `bug-filing` skill instead. Requires a Bugzilla API key.
Help prepare a Firefox security approval request by analyzing local commits/changes and drafting answers to the sec-approval questionnaire. Use when setting sec-approval? on a Bugzilla bug.
Help prepare a Firefox uplift approval request (Beta, Release, and/or ESR) by checking whether patches are already on the bug, auditing sanitization once for sec-* bugs, and drafting the approval comment per https://wiki.mozilla.org/Release_Management/Uplift_rules. Use after a fix is ready and needs to ride into a stabilization branch. May or may not follow sec-approval (sec-moderate typically skips sec-approval but may still need uplift).
Reference for constructing revision-pinned source, spec, and bug permalinks (Searchfox, GitHub, GitLab, googlesource, Codeberg, Chromium, FFmpeg, Bugzilla, specs). Use when citing code, specs, or bugs in analysis or reports and you need stable, revision-pinned URLs instead of trunk/HEAD links.
Firefox bug triage assistant for media, web conferencing, and graphics related issues.
Generate technical documentation for a specified technology domain with optional customization.
| name | bugzilla-wrangler |
| description | Support deep dive analysis of Bugzilla reports within specific Products and Components. |
Discover and surface recurring themes within bug data in your Products and Components. Connect disparate reports to shared themes, build timelines, assign signal strength, and present actionable findings.
The core logic of weighting is covered in Signal Scoring. You can tweak the ranking system there to alter what you find. Currently we're finding major and minor themes buried in the noise of bugzilla. The tool will surface existing issues, and for more current date ranges, emerging themes that haven't yet been widely recognized.
Typical use:
/bugzilla-wrangler (media, last three months)
/bugzilla-wrangler web-conferencing date range:10-01-2025 to 03-18-2026
/bugzilla-wrangler custom:Core::Audio/Video, Core::WebRTC: Signaling, Firefox for Android::Media
At invocation, resolve these two settings before doing anything else. State both in the Session Info section of the report.
Scope profile — If the user passed a custom: prefix (e.g. custom:Core::Audio/Video, Firefox for Android::Media), parse the value as described in the custom scope profile below. Otherwise, if the user specified a named platform or topic (e.g. "android", "web-conferencing", "graphics"), use that profile from the Scope Profiles section. If neither, use media.
Seed date window — If the user specified a date range (e.g. "Q1 2025", "last 90 days", "January through March 2026"), parse it into an explicit YYYY-MM-DD lower bound and upper bound and use those in Step 1. Otherwise use a default window from the last three months.
Seed mode — Use the hybrid seed strategy in Step 1 by default (recently-created bugs fill up to 75% of seed slots; recently-changed bugs fill the remainder). If the user passed --no-mixed-seed, use the creation-time-only strategy instead. State the active seed mode in Session Info.
At invocation, check whether the user specified a platform or topic (e.g. "android", "web-conferencing", "graphics"). If so, use the matching profile below. If not, use media.
Defined inline at invocation. The value is a comma-separated list of Product::Component pairs. Both :: and : are accepted as the product/component delimiter (e.g. Core::Audio/Video or Core:Audio/Video). Example invocation:
/bugzilla-wrangler custom:Core::Audio/Video, Core::WebRTC: Signaling, Firefox for Android::Media
Parsing rules:
, to get individual pairs.:: or : to separate the product (left side) from the component (right side). Strip whitespace from both sides.Core and emit a visible warning in Session Info.The resolved product/component set is used identically to a named scope profile for all subsequent steps. Use custom as the scope name in cache filenames and report headers.
Breadcrumb — Any resource associated with a theme: a bug report, test case, crash report, etc.
Signal — A measure of theme importance based on the number and quality of breadcrumbs. See scoring rules below.
Use Mozilla's Bugzilla REST API directly through WebFetch. Do not use the moz MCP server for Bugzilla access.
Parse ../shared/Bugzilla.md for details on how to access and use Bugzilla data effectively during this triage process. Check for updates, this skill is a critical part of our workflow.
reports/wrangler_cache_<scope>.json (e.g. wrangler_cache_android.json, wrangler_cache_media.json, wrangler_cache_custom.json). Never read or write a different scope's cache file.
?id=A,B,C rather than one request per bug.id, summary, assignee, component, severity, status, creation_time, last_change_time, keywords, depends_on, blocks, cc_countcc_count separately if missing. The field can be omitted by the API depending on visibility. If cc_count is absent after a batch fetch, retrieve it with a targeted request: ?id=A,B,C&include_fields=id,cc_count. Always persist cc_count in the cache so signal scoring can use it without a second fetch.Apply these rules when ranking themes. Higher-weight factors are listed first.
| Factor | Effect |
|---|---|
| High duplicate count or independent reporters with near-identical summaries | Both indicate broad user impact; weight more heavily when duplicates originate from unrelated reporters rather than a single coordinated source (e.g. a forum post) |
Meta/tracking bug exists (keywords=meta) | Strong multiplier — use its total dependency count as a proxy for theme breadth |
cc_count ≥ 10: moderate boost; cc_count ≥ 25: strong boost | Proxy for how many users are tracking the issue |
comment_count ≥ 15: moderate boost; comment_count ≥ 40: strong boost | Proxy for community demand and sustained engagement around the issue. Add comment_count to include_fields. |
| Bug severity is S1 or S2 | Validated internally as a major issue. |
| Bug priority is P1 or P2 | Validated internally as a priority to fix. |
| Bug status is NEW or ASSIGNED | Counts more than UNCONFIRMED (triage has validated it) |
| Crash volume and trend | >1 000 reports in 30 days is a strong signal; a rising trend (comparing two 30-day windows via Socorro) elevates the theme regardless of other factors |
regression keyword present | Higher priority — represents a user-visible regression from a prior release |
| Non-Mozilla commenter activity | A thread dominated by non-mozilla.com accounts signals direct end-user impact rather than internal triage noise; weight proportionally to the number of distinct non-Mozilla participants |
last_change_time older than 30 days with no resolution | Negative factor: marks a theme as stagnant — call this out in the report rather than using it as a positive signal |
stalled keyword with S1 or S2 severity | Notable: acknowledged this needs to be fixed but there are no engineers available |
A theme requires at least 3 breadcrumbs to be included in the report.
Before starting, identify the active scope profile. If the user specified a platform (e.g. "android") at invocation, use that profile's Product and Component set for all seed searches and keyword expansion queries. Otherwise use media. State the active profile in the Session Info section of the report.
Retrieve a seed sample of up to 50 bugs using the seed date window resolved in Run Configuration.
The common filter parameters below apply to all queries in this step:
status=UNCONFIRMED&status=NEW&status=ASSIGNED&status=REOPENEDf2=bug_severity&o2=notequals&v2=-- and f3=bug_severity&o3=notequals&v3=N/Af4=keywords&o4=notsubstring&v4=intermittent-testcase — exclude CI-only intermittent bugsorder=changeddate%20DESCDefault seed mode (no --no-mixed-seed flag):
Run two queries in parallel:
Query A — recently created (up to 38 bugs):
creation_time=<lower-bound>f1=creation_ts&o1=lessthan&v1=<upper-bound>limit=38Query B — recently changed (up to 25 candidates):
f1=delta_ts&o1=greaterthan&v1=<lower-bound>f2=delta_ts&o2=lessthan&v2=<upper-bound>limit=25Merge the two result sets: start with all bugs from Query A, then append bugs from Query B whose IDs are not already present, until the combined list reaches 50. Discard any remainder from Query B. Record the final composition as <A-count> created + <B-unique-count> changed = <total> unique for the Session Info section.
Creation-time-only mode (--no-mixed-seed flag):
Fetch up to 50 bugs created within the seed date window:
creation_time=<lower-bound>f1=creation_ts&o1=lessthan&v1=<upper-bound>limit=50Save results to the cache.
Low seed count check: After saving, count the results. If fewer than 10 bugs were returned, emit a visible inline warning before proceeding — for example:
Warning: Low seed count — only N bugs returned for the
<scope>profile over<window>. Results may not be representative. Consider widening the date window or relaxing filters.
Continue with the analysis regardless, but carry the low-seed flag into the report.
Read the seed bugs and group them by shared symptoms, components, keywords, or user-facing impact. Name each candidate theme descriptively (e.g., "Bluetooth audio stutter", "WebRender visual glitches").
For each seed bug, collect depends_on and blocks IDs. Batch-fetch them in a single request. Repeat up to depth 5. Do not restrict your search to specific bugzilla products and components. Search any Product or Component when crawling for breadcrumbs.
Prioritize bugs with keywords=meta — these tracking bugs directly amplify signal. Note the total dependency count of any meta-bug you find.
When crawling deps of a meta-bug that has more than 50 dependencies, do not fetch all of them. Instead, filter to the most recent by sorting: use order=changeddate%20DESC&limit=20 to fetch only the 20 most recently changed deps. This avoids context overload from omnibus tracking bugs (e.g. site-scout meta).
Alongside the dep crawl, for any seed bug or dep bug (at any crawl depth) that carries the crash keyword, cross-reference with Socorro to quantify user impact. Use socorro-cli to look up each crash signature:
socorro-cli search --signature "TrackBuffersManager::RemoveFrames"
socorro-cli search --signature "CAudioClient::GetCurrentPadding"
Android/Fenix scope: Java and Kotlin stack traces use a different product namespace in Socorro and return 0 results when searched without a product filter. For the android scope profile, always pass --product Fenix and use --facet signature to discover relevant crash signatures rather than guessing Java class names directly:
socorro-cli search --product Fenix --facet signature --num-facets 20
socorro-cli search --product Fenix --facet signature --query "media" --num-facets 20
Pick the most relevant signatures from the facet output, then drill into each with a standard --signature search to get volume and trend data.
For each crash bug, note:
--date to compare two windows (e.g. --date ">=2026-02-13" vs --date ">=2026-01-14") to determine directionAttach these metrics to the bug record so signal scoring in Step 5 can use them immediately. A crash with >1 000 reports in 30 days should elevate the theme's rank regardless of other signals. Add a Crash Volume (30d) column to affected themes' breadcrumb tables in the report.
If socorro-cli is unavailable, note that enrichment was skipped and recommend it as a follow-up action in the closing section.
For each developing theme, run 1–2 targeted searches against the same component set to find bugs not connected via the dependency graph. Examples:
keywords=regression in Graphics: WebRenderbluetooth or audio sink in Audio/Videokeywords=webcompat:site-report search across the Audio/Video component set; site-report bugs (e.g. a specific site's audio broken) rarely appear in dep graphs but carry strong user-impact signal and are frequently filed independently by webcompat triage.Add any relevant results to the theme's breadcrumb list.
Apply the signal scoring rules to each theme. Rank themes by total signal. Discard any theme with fewer than 3 breadcrumbs. Independently, choose a few themes to highlight which might represent emerging issues, even if the initial signal for the theme is weaker.
Present a short summary of the themes you've found. Save the full research report under "./reports". The filename format should be wrangler-<scope>-<DATE>.md. If the file already exists, we should not overwrite it, but instead create a new file with a suffix, like wrangler-<scope>-<DATE>-2.md, and so on.
Session Info:
mixed-seed (<A> created + <B> changed = <total> unique) or creation-time-only (--no-mixed-seed)Seed Info:
Per theme:
Bug ID | Summary | Severity | Status | Last ChangedClosing:
Format Notes