* improve: enhance computer-vision-engineer based on automated review
- Fix deprecated TensorRT 10.x API (build_serialized_network, set_memory_pool_limit)
- Add Foundation & Zero-Shot Vision Models section (SAM2/SAM3, Grounding DINO,
CLIP, Florence-2, multimodal LLMs) with a model selection framework
- Update default detector to YOLO11 with AGPL-3.0 licensing note and
RT-DETRv2/RF-DETR alternatives
- Modernize face recognition guidance to InsightFace/DeepFace, keep dlib
face_recognition as a lightweight fallback only
- Add multi-object tracking (ByteTrack via model.track) to video analysis
- Replace unimplemented document-structure stub methods with a layoutparser-based
approach and pointers to unstructured/VLM alternatives
- Add a Compliance & Ethical Considerations section for biometric/PII data
- Align frontmatter/structure with sibling data-ai agents: add model: sonnet,
Glob/Grep tools, multi-example description, and an Integration with Other
Agents section with explicit boundaries vs ai-engineer and ml-engineer
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: reset ByteTrack state between independent video streams
Reusing a VideoAnalyzer instance for a new stream would otherwise carry
stale track IDs from the previous session, since persist=True caches
tracker state on the model's predictor.
Addresses review feedback on PR #821.
* fix: stop prior stream's threads before starting a new one
start_real_time_analysis previously only reset tracker state but left
any earlier capture/process threads running against the same
frame_queue/result_queue/model, so a reused VideoAnalyzer instance could
mix frames and tracker state across streams. Add stop_processing() to
join the old threads and drain the queues before starting a new stream.
Addresses further review feedback on PR #821.
* fix: correct layoutparser color channel and text-region filter
- Convert BGR (cv2.imread) to RGB before Detectron2LayoutModel.detect(),
since PubLayNet was trained on RGB
- Include PubLayNet's 'Title' and 'List' labels in text_regions, not
just 'Text', so headings and list blocks aren't dropped
Addresses cubic review feedback on PR #821.
* fix: unblock capture thread on stop and fail loudly if it won't die
cap.read() blocks until a frame arrives, so join(timeout=5.0) alone
could return with the capture thread still alive and start_real_time_analysis
would then reuse the shared model/queues underneath it. stop_processing
now releases the VideoCapture to unblock read() immediately, and raises
if a thread still hasn't exited after the join timeout instead of
silently proceeding.
Addresses further greptile review feedback on PR #821.
* fix: re-check thread liveness on retry instead of the processing flag
stop_processing() sets self.processing = False before it can guarantee
the old threads have exited (e.g. one is still blocked in model.track()
past the join timeout and it raises). A caller retrying
start_real_time_analysis after that RuntimeError would see
self.processing already False and skip straight to relaunch, letting
the still-alive worker resume against the shared model/queues once it
unblocks. Guard on actual thread.is_alive() instead so a retry keeps
re-joining until the old threads are confirmed dead.
Addresses further greptile review feedback on PR #821.
* fix: call stop_processing unconditionally to always drain queues
Gating stop_processing() on thread liveness skipped the queue drain
whenever the previous call's stuck thread had exited by the time of a
retry (the liveness check would then be False), leaking stale
frames/results from the old stream into the new one. stop_processing is
idempotent for dead/None threads, so call it unconditionally instead.
Addresses cubic review feedback on PR #821.
* fix: guard tracker reset against predictor without trackers attribute
Ultralytics only registers .trackers on the predictor once model.track()
has run at least once. If this model instance was previously used with
model.predict()/model() (plain inference, no tracking), self.model.predictor
exists but lacks a trackers attribute, so the direct attribute access
raised AttributeError and aborted stream startup. Use getattr with a
default instead of assuming the attribute exists.
Addresses further greptile review feedback on PR #821.
---------
Co-authored-by: Claude <noreply@anthropic.com>
Regenerate docs/components.json and dashboard public artifacts to pick up
new components (livetennisapi MCP, open-banking-io skill) and refreshed
download stats.
- doordash-group-orders: quote description (contains ": ")
- ray-data, ray-train: quote `ray[data]` / `ray[train]` inside the
dependencies flow sequence (nested brackets are invalid there)
- motion-canvas: quote `@motion-canvas/*` dependency entries (`@` is a
reserved indicator and cannot start a plain scalar)
- agirails-agent-payments: drop a stray leading `+` on the description
line left over from a diff paste
- scholar-evaluation: add missing frontmatter (name + description taken
from the existing Overview paragraph)
All six were skipped by `npx skills add davila7/claude-code-templates`;
verified 0 skipped after this change.
Co-authored-by: cj <cj-ant@users.noreply.github.com>
* improve: enhance data-analyst based on automated review
- Add missing model field (sonnet)
- Add clarifying-questions preamble before starting analysis
- Fix stale Data Studio reference to Looker Studio
- Ground vague concept lists with concrete tools & libraries
- Add Responsible Analysis section for data privacy/governance
- Remove stray duplicate blank line
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: address cubic review findings on data-analyst
- Remove orphaned Communication Protocol / Analysis Context block
(get_analysis_context) left dangling after the "Query context manager"
step was replaced by the clarifying-questions preamble; the reference
sibling data-scientist.md has no equivalent block either.
- Drop the duplicated tool-name parenthetical on "Business intelligence"
now that the same names live in the single "Tools & libraries" list.
Co-Authored-By: Claude Code <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
There is no cost or FinOps MCP in the catalog today. nable answers spend
questions across AWS, Azure, GCP, Kubernetes and 15+ SaaS and AI
providers, and drafts the Terraform fix as a pull request a human
reviews.
Runs with uvx, like the nine existing uvx components. Needs no API key
in the config: it reads the cloud credentials already on the machine.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* improve: enhance accessibility-tester agent based on automated review
- Fix pa11y guidance to use --runner axe (default htmlcs runner is WCAG 2.0-era, misses 2.1/2.2 violations)
- Note axe-core version requirement (>=4.5) for reliable WCAG 2.2 rule coverage
- Annotate which WCAG 2.2 new criteria are actually automatable (only 2.5.8 via axe-core target-size)
- Add a fully worked example finding to anchor severity/remediation phrasing
- Add document/PDF accessibility checklist item (Section 508/EAA scope)
- Add forced-colors / prefers-contrast manual checklist item
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: cover prefers-contrast alongside forced-colors in accessibility-tester
Addresses review feedback on PR #814 — the manual checklist tested
Windows High Contrast (forced-colors) but omitted the distinct
prefers-contrast media feature, which interfaces can also use for
alternate high-contrast styling.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QTTuf1dEt7QW83eyAPV7Lm
* fix: require both forced-colors and prefers-contrast checks explicitly
The previous wording used "or", allowing an audit to satisfy the
checklist item by testing only one of the two distinct contrast
display modes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QTTuf1dEt7QW83eyAPV7Lm
* fix: correct pa11y WCAG 2.2 coverage claim and axe-core version check scope
pa11y's axe runner maps WCAG2AA to wcag2a/wcag21a/wcag2aa/wcag21aa tags
only (confirmed against pa11y/pa11y#666 and lib/runners/axe.js) — it does
not include wcag22aa, and pa11y has no --tags CLI flag. Also scope the
axe-core version check to the tool actually being run, since
@axe-core/cli --version only reports its own bundled axe-core, not
pa11y's or @axe-core/playwright's independently-resolved versions.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QTTuf1dEt7QW83eyAPV7Lm
* fix: give a concrete command for checking per-tool axe-core versions
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QTTuf1dEt7QW83eyAPV7Lm
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance model-evaluator based on automated review
- De-anchor hardcoded/aging model names (GPT-4o, Gemini 1.5, etc.) in
frontmatter examples and Model Categories, replacing them with
capability-tier language and a prominent instruction to verify
current model IDs via WebSearch/WebFetch before recommending or
testing
- Add a "Required Initial Step: Requirements Gathering" section,
matching the pattern used by sibling agents llm-architect and
prompt-engineer
- Add WebFetch to the tools list for pulling full pricing/leaderboard
pages (LMSYS Arena, HELM) beyond WebSearch snippets
- Add Inspect AI and OpenAI Evals to the Standard Frameworks & Tools
table
- Fix imprecise terminology: replace "GitHub Copilot" (a product, not
a benchmarkable model) with actual code-specialized models, and
clarify "Vision" models are a native capability rather than a
separate SKU
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: align tier language in model-evaluator example
The first description example's Context named balanced/flagship tiers
but the assistant response evaluated budget/balanced tiers instead.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FHagP9k8UdbMyPJHjmYaTy
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance prompt-engineer based on automated review
- Add WebSearch to tools frontmatter, matching sibling ai-specialists agents
- Add 4th description example covering Claude Code subagent/skill prompt optimization
- Add guidance on avoiding aggressive imperative language (CRITICAL/YOU MUST/NEVER EVER) that causes overtriggering on current models
- Route target-model answer to model-specific prompting guidance (verbosity, effort calibration, delegation, tool-triggering)
- Add long-context document placement guidance (documents near top, instructions/query at end)
- Expand agentic prompting into concrete sub-guidance (tool-triggering, overeagerness containment, destructive-action confirmation, tool-error handling, autonomy boundary)
- Sharpen budget_tokens deprecation language (returns HTTP 400 on current models; effort/max_tokens is the replacement)
- Reference Structured Outputs as a response-parsing option
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: qualify budget_tokens claim by model generation, regenerate catalog
Addresses PR review feedback:
- cubic-dev-ai/greptile: dashboard/catalog artifacts were stale after the
prompt-engineer edits (old tool allowlist, 3-example description) —
regenerated via scripts/generate_components_json.py.
- cubic-dev-ai: the budget_tokens deprecation claim was overbroad — it now
varies correctly by model generation instead of asserting a blanket
HTTP 400 on all "current" models.
Co-Authored-By: Claude Code <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance task-decomposition-expert based on automated review
- Trim tools allowlist to Read, Glob, Grep, WebSearch (drop unused Write/Edit/Bash)
- Add "Boundaries with Related Agents" section distinguishing from project-manager and scrum-master
- Instruct grounding the WBS in the actual codebase via Read/Glob/Grep before finalizing
- Add PERT (three-point) estimation guidance for Medium+ complexity tasks
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: address review feedback on task-decomposition-expert
- Regenerate the dashboard content mirror so it matches the trimmed
tool allowlist and new prompt sections (Greptile/cubic P2).
- Reword the Boundaries section so it no longer claims project-manager
and scrum-master appear in the Agent Handoff Plan table (cubic P3).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZwwQ2MD4wsrwTJi5yM515
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance salesforce-expert agent based on automated review
- Add required model field (sonnet) to frontmatter
- Replace invalid tools list (vscode, execute, sfdx-mcp/*, agent, todo) with valid Claude Code tool names
- Replace generic description with routing-oriented description plus 3 <example>/<commentary> blocks
- Add Agentforce & Agent Actions subsection (exposing Apex as Agentforce Actions, Topics/Instructions, guardrails)
- Add Flow vs. Apex decision-guidance subsection
- Add Lightning Web Security (LWS) and API-version-pinning guidance to operationalize the Release Aware Developer persona
- Add Salesforce CLI Commands Reference section using the current unified sf CLI, noting sfdx is legacy
- Normalize bullet indentation to a single "- " marker
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: remove TodoWrite from salesforce-expert tools allowlist
TodoWrite isn't in StructuralValidator's accepted tool identifiers
(Read, Write, Edit, Bash, Glob, Grep, WebSearch, WebFetch, *), which
was flagging STRUCT_W006 on the PR.
---------
Co-authored-by: Claude <noreply@anthropic.com>
- Add required model field (sonnet) to frontmatter
- Name specific database engines (PostgreSQL, MySQL, MongoDB, Redis) in description and Focus Areas
- Add credential-safety instruction to Approach (env vars / secrets manager)
- Add Glob, Grep to tools field for locating existing config files
- Add differentiation/handoff note vs postgres-pro and database-optimizer
- Name concrete per-engine monitoring commands and schema migration tooling
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
Co-authored-by: Claude <noreply@anthropic.com>
Local MCP connects Claude Code to 160+ native macOS apps (Mail,
Calendar, Contacts, Messages, Notes, Reminders, Safari, Teams,
Slack, WhatsApp, OneDrive, Google Drive, M365, Outlook, Office
docs, and more) via a free local tray app. No API keys required.
https://local-mcp.com
Co-authored-by: LMCP <lmcp@LMCPs-Mac-mini.local>
- Resolve name collision: rename cli-tool/components/agents/deep-research-team/search-specialist.md
to multi-source-searcher.md (both filename and frontmatter name field), since it
duplicated the canonical, higher-adoption ai-specialists/search-specialist.md and
both installed to the same .claude/agents/search-specialist.md target path.
Updated the cross-reference in research-analyst.md accordingly.
- Tighten WebFetch guidance in ai-specialists/search-specialist.md to request
verbatim extraction and note that WebFetch only returns raw text unmodified for
Markdown sources under ~100K characters (otherwise a smaller model paraphrases it).
- Add a "Handling Untrusted Content" section instructing the agent to treat fetched
page content as untrusted data, not instructions, and flag rather than act on
embedded directive-like text (prompt-injection hardening).
- Clarify that allowed_domains/blocked_domains are mutually exclusive per WebSearch call.
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance llms-maintainer based on automated review
- Make content-discovery directories framework-aware (Astro/Nuxt/Next/SvelteKit) instead of a fixed generic list
- Add detection and scan paths for Docusaurus, VitePress, and Jekyll
- Raise maxTurns from 20 to 50 and add a batching/graceful-truncation strategy for large sites
- Add "Use PROACTIVELY" to the description for reliable automatic delegation
- Gate automatic git commits behind explicit user confirmation, consistent with the deletion-confirmation safety rule
- Add optional llms-full.txt companion generation, opt-in only
- Clarify that description/title length limits and the blockquote/H2 sections are house-style conventions, not llms.txt spec requirements
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: use detected output path when staging llms.txt, not hard-coded public/llms.txt
Addresses Greptile review feedback on PR #794: the Step 8 git-add command
was hard-coded to public/llms.txt, which misses the actual output path for
SvelteKit/Hugo/Docusaurus (static/llms.txt), VitePress (<srcDir>/public/llms.txt),
Jekyll (llms.txt at repo root), and the optional llms-full.txt companion.
Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: resolve Next.js hybrid-router ambiguity and Jekyll _posts/ ignore-rule conflict
Addresses cubic review feedback on PR #794:
- The two separate Next.js detection bullets were not mutually exclusive;
a hybrid app/+pages/ setup would have legacy pages/ routes silently
omitted. Now scans both when both exist, preferring app/ for output.
- Step 3's blanket "/_*" ignore rule would have silently skipped every
Jekyll post, since _posts/ starts with an underscore. Added an explicit
exception for _posts/ while still ignoring other Jekyll internals.
Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: extend Jekyll ignore-rule exception to configured collection dirs
Addresses cubic review feedback on PR #794: the underscore-path exception
only covered _posts/, but Step 1 also tells the agent to scan Jekyll's
configured collections (e.g. _projects/, _team/), which are similarly
underscore-prefixed and were getting silently filtered by Step 3's ignore
rule.
Co-Authored-By: Claude Code <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance sales-engineer based on automated review
- Add required `model: sonnet` field to frontmatter
- Replace vestigial "context manager" pattern with explicit "When Invoked" user-ask step
- Remove fabricated metrics (demo/win-rate figures); use session-grounded placeholders instead
- Add Human-in-the-Loop Pause Criteria for pricing, SLA, compliance, and POC sign-off
- Add Source Boundaries for Research (public sources only, citations, no pretexting)
- Add Anti-Fabrication section for compliance certifications, benchmarks, SLAs
- Consolidate redundant bullet sections into a single Core Practices section
- Modernize qualification methodology to MEDDIC/MEDDPICC with BANT as secondary reference
- Add time-boxed POC guidance with upfront success criteria and decision gates
- Add scope-boundary note to description (hand off pricing/contracts and compliance drafting)
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* address review: avoid redundant discovery questions when context is already supplied
Per Greptile review comment on PR #791.
* address review: drop dangling "(below)" cross-reference in Anti-Fabrication section
Per cubic review comment on PR #791.
---------
Co-authored-by: Claude <noreply@anthropic.com>
Dani