- 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>
* improve: enhance hackathon-ai-strategist agent based on automated review
- Expand tools to Glob, Grep, Write for codebase exploration and saving deliverables
- Instruct WebSearch/WebFetch use for current model releases, sponsor API changes, and recent winning techniques
- Add Boundaries with Related Agents / Integration with Other Agents sections
- Merge duplicated Strategic Guidance / Optimizing for Constraints sections
- Add AI-assisted rapid-prototyping tools (Lovable, Bolt.new, v0, Replit Agent) to strategic guidance
- Add submission-format context gathering question and cross-reference in Phase 5
- Instruct confirming actual published judging rubric before defaulting to fixed weights
- Add solo-hacker and virtual/remote-hackathon callouts
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: add fallback when submission-format details are unavailable
Addresses review feedback on PR #790: the new context-gathering
question about submission platform/format could stall the agent's
advice when the event hasn't published those details yet. Fall back
to a conservative default (~3 min cap, public repo) and flag it as
an assumption to confirm later, instead of blocking.
* fix: scale pitch outline to the actual submission cap
Addresses cubic review feedback on PR #790: Phase 5 told the agent
not to assume a 3-minute cap, but the only pitch outline provided
was hard-coded to exactly 3 minutes with no scaling guidance. Reframe
it as a 3-minute baseline with an explicit proportional-scaling rule.
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance api-architect agent based on automated review
- Add idempotency-key guidance (Idempotency-Key header) for retry/backoff on non-idempotent methods
- Update auth guidance from OAuth 2.0 to OAuth 2.1 (PKCE S256, Client Credentials, no Implicit/ROPC)
- Resolve REST/GraphQL/gRPC scope inconsistency (gRPC recommendations hand off to api-designer)
- Adopt RFC 9457 Problem Details as the REST error contract
- Reference standard rate-limit headers (RateLimit, RateLimit-Policy, Retry-After) in resilience/security guidance
- Add cursor/offset pagination guidance for GET-all
- Add OpenTelemetry tracing guidance for the resilience layer
- Add Deliverables, Bash Usage Constraint, and Integration with Other Agents sections (mirroring api-designer.md conventions)
- Scope Bash tool access to output validation (redocly lint, graphql-inspector validate)
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: address PR review findings in api-architect
- Add PATCH to the REST methods intake list — the new idempotency
default already referenced POST/PATCH but PATCH wasn't selectable
- Fix rate-limit header mismatch: client-side backoff now parses the
RateLimit header's reset parameter instead of a nonexistent
RateLimit-Reset header, matching what the security checklist tells
servers to emit
Addresses review-bot findings on PR #789.
* fix: correct RateLimit header parameter name to t
draft-ietf-httpapi-ratelimit-headers-11 uses the t parameter
(effective window in seconds) on the RateLimit header, not a
"reset" parameter. Addresses greptile-apps follow-up finding on
PR #789.
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance ml-engineer based on automated review
- Modernize tooling ecosystem with named, current tools (KServe, vLLM/Triton, Feast/Tecton/Hopsworks, Evidently/WhyLabs/Arize) and note Seldon Core v2's BSL license
- Ground bare "Data validation" bullets in named tools (Great Expectations, Pandera)
- Add scope-clarifying sentence distinguishing this agent from machine-learning-engineer and mlops-engineer
- Reframe hardcoded performance targets as illustrative/configurable SLAs
- Name fairness/explainability tooling (Fairlearn, Aequitas, SHAP, LIME) and add model card/regulatory awareness note
- Add ML-specific security bullets: artifact integrity, training data PII/leakage checks, adversarial robustness testing
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: move ml-engineer scope boundaries into frontmatter description
Agent auto-selection reads the frontmatter description, not the body
prompt text, so the scope-clarifying sentence added in the previous
commit had no effect on delegation. Move it into the description
field where it can actually reduce ambiguity with machine-learning-
engineer, mlops-engineer, and ai-engineer.
Addresses review feedback from greptile-apps[bot] on PR #787.
* fix: route prompt-text optimization to prompt-engineer, not ai-engineer
ai-engineer's own description explicitly hands off prompt-text-only
optimization on an already-chosen model to prompt-engineer (ai-specialists
category). ml-engineer's scope note was steering all "prompting work" to
the generalist ai-engineer instead, which could reintroduce the routing
ambiguity this review is meant to reduce.
Addresses review feedback from cubic-dev-ai[bot] on PR #787.
* fix: resolve contradiction between ml-engineer's example and its own delegation boundary
Example 2 in the frontmatter description depicted ml-engineer handling
pure inference-serving optimization (quantization, serving-strategy
comparison, canary rollout) - exactly the work the new scope note says
to delegate to machine-learning-engineer. Rewrote the example so
ml-engineer owns the training-pipeline root cause (feature drift,
retraining) and explicitly hands off the serving-side optimization,
following the same hand-off pattern already used in ai-engineer.md's
examples.
Addresses review feedback from greptile-apps[bot] on PR #787.
* clarify: distinguish model-level lifecycle work from platform infra in Example 3
Sharpens Example 3's commentary to explicitly frame deployment/
monitoring/retraining as model-level lifecycle ownership (a specific
model's own rollout and retraining loop), distinct from the underlying
platform/infrastructure automation (CI/CD, GPU orchestration, cross-model
versioning) that mlops-engineer.md's own description already claims.
Full disambiguation between the three sibling agents' descriptions would
require editing mlops-engineer.md and machine-learning-engineer.md too,
which is out of scope for this single-file review.
Addresses review feedback from greptile-apps[bot] on PR #787.
* fix: remove ambiguous "performance optimization" from opening invocation trigger
The opening sentence of the frontmatter description is the first thing
agent selection matches against, and it still claimed raw "performance
optimization" as ml-engineer's trigger even though the same description
delegates deep inference-serving optimization to machine-learning-engineer
later on. Reworded the opening trigger to fold the redirect in directly,
so a pure serving-latency-optimization request no longer matches
ml-engineer's own invocation criteria.
Addresses review feedback from greptile-apps[bot] on PR #787.
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance data-scientist based on automated review
- Replace pinned model claude-sonnet-4-5 with alias sonnet
- Add WebSearch to tools (needed for literature review/methodology selection tasks already in the prompt)
- Name concrete libraries: Fairlearn/AIF360 (bias/fairness), SHAP/LIME (interpretation), DoWhy/EconML/CausalML (causal inference), Optuna/Ray Tune/Hyperopt (hyperparameter tuning), DVC (versioning)
- Tighten data science checklist to measurable/testable criteria instead of vague adverbial items
- Upgrade pre-analysis clarifying questions into a structured, numbered gate that blocks proposing a model/methodology until key unknowns are resolved
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: repair double-escaped newlines in data-scientist description
Pre-existing bug where the description field used \\n instead of \n,
rendering as literal backslash-n text in the agent picker/dashboard.
Also tightens grammar in the pre-analysis gate added in the prior commit.
* fix: scope modeling requirements to modeling tasks in data-scientist gate
Addresses Greptile review feedback: the clarification gate and checklist
required a success-metric threshold and held-out-test/cross-validation
steps even for pure exploratory profiling, visualization, and clustering
work the agent explicitly supports.
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance droid agent based on automated review
- Fix frontmatter to use valid Claude Code subagent conventions (tools: Read, Grep, Glob, Bash, WebFetch, WebSearch instead of GitHub Copilot aliases; model: sonnet instead of pinned dated ID)
- Remove GitHub Copilot Integration section (contradicted actual .claude/agents/ install path) and replace with a Claude Code usage section
- Fix outdated --enabled-tools flag to current --restrict-tools/--additional-tools/--disabled-tools
- Remove stale hardcoded model IDs from droid exec --model examples, point to docs.factory.ai/models instead
- Add missing droid exec flags: --fork, -w/--worktree, --use-spec, -r/--reasoning-effort
- Add <example>/<commentary> blocks to the description frontmatter matching sibling agent conventions
- Clarify autonomy tier boundaries (medium = no push, high = push/deploy with safety checks)
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: remove dangling model-ID hint from GitHub Actions example
The parenthetical referenced a --model flag that was removed from
the command in an earlier edit, leaving a confusing dangling hint.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01B3THSCtx7huxkRFmmUXne7
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance llm-architect based on automated review
- Add Glob and Grep to tools list (parity with sibling agents)
- Add SGLang to serving framework selection; remove stale vLLM version gates
- Add Anthropic Contextual Retrieval technique and Voyage AI embedding option
- Add GraphRAG/agentic retrieval escalation callout for multi-hop queries
- Add Unsloth as fine-tuning library option for LoRA/QLoRA
- Add ORPO and KTO preference-alignment methods to method-selection table
- Add "verify current model IDs" disclaimer to multi-model routing section
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: clarify ORPO vs DPO/KTO selection criterion in llm-architect
Both rows now key off dataset format (paired vs unpaired preferences)
instead of mixing in GPU count, so ORPO no longer overlaps with the
adjacent DPO/KTO rows.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Eckv6ssk3XuRXpbAzCxPgA
* fix: split DPO and GRPO into separate rows with correct prerequisites
GRPO is a reward/verifier-driven RL method, not a paired-preference
method — the previous merge wrongly implied it needs paired preference
data and a prior SFT checkpoint like DPO does.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Eckv6ssk3XuRXpbAzCxPgA
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Update retired model ids in settings components and docs
* Fix CLI default and sync generated dashboard copies
Two gaps from the first commit:
1. evaluation.py only had the function signature default updated. The
argparse `--model` default still pointed at the retired model, and
since `args.model` is always passed explicitly to run_evaluation, that
CLI value overrode the new default. Anyone running the script without
-m still got the retired model, so the change had no practical effect.
The usage example and reference/evaluation.md carried the same stale
ID.
2. The settings components under cli-tool/ were updated but their
generated copies under dashboard/public/component-content/ were not,
so the public dashboard would keep serving the retired IDs. Copies are
now regenerated from their sources rather than hand-edited.
---------
Co-authored-by: kasparovabi <kasparovabi@users.noreply.github.com>
* Add avoid-ai-writing skill (productivity)
Vendors the SKILL.md from conorbronsdon/avoid-ai-writing at tag v3.22.2
(MIT) plus its licence. The skill audits and rewrites prose to remove AI
writing patterns, with detect / rewrite / edit-in-place modes and optional
voice and context profiles.
Everything below the frontmatter is byte-identical to upstream. The
frontmatter is reshaped for this repo's catalog: author promoted from
metadata, tags promoted and rewritten into the bracketed list
generate_components_json.py can parse, repo added, and the openclaw emoji
and compatibility lines dropped.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Remove the three upstream-only references from the vendored skill
All three resolved nowhere in this repo, so a user who installs the skill
here gets instructions that fail or point at nothing:
- "until **this repo** measures the ratios itself" -- vendored here,
"this repo" reads as claude-code-templates, which is not what measures
anything. Now states the condition without the repo-relative pointer.
- "(see `detector/CATEGORIES.md` §C)" -- that file does not exist here.
The sentence keeps its point: the structural detector for that rule was
reverted, which is why the tolerance matrix is the wrong home for it.
- The optional `node detector/validate.js` block -- the engine does not
ship in this repo, and the "If the repo ships the detector engine" guard
covered the command but not the surrounding claim. Replaced with the
same checks written as a manual verification step, so edit mode keeps
the preservation contract without depending on a binary that is not
here.
This gives up byte-identity with upstream v3.22.2 below the frontmatter,
which the PR description claimed. A future re-sync now needs these three
spans reapplied rather than a straight copy. That is the right trade: the
byte-identity was a maintenance convenience for me, and a pointer that
goes nowhere is a defect for whoever installs this.
Catalog parse re-checked after the edit -- version, author, repo,
license, keywords, and references all still resolve, no empty fields.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Re-vendor from upstream v3.22.3: five prose-contract clarifications
Upstream applied the five applicable findings from cubic's review of
this PR (conorbronsdon/avoid-ai-writing#97, released as v3.22.3):
tables join the edit-mode flag-don't-fix list, the rewrite job line
scopes 'all AI-isms removed' to editable ones, edit mode gets an
explicit instruction boundary, the tracking-parameter fix leaves the
rest of the query string alone, and the second-pass audit must say
when its text supersedes section 2. Same vendoring transform as
before: frontmatter rewrite plus the three documented body spans.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
* improve: enhance ai-engineer based on automated review
- Reposition scope as generalist for both classical ML and generative-AI/LLM engineering, with explicit handoffs to llm-architect, ml-engineer/machine-learning-engineer, and prompt-engineer
- Add Generative AI / LLM Engineering section (LLM API providers, RAG basics, agentic patterns, fine-tuning, evaluation frameworks)
- Add model: sonnet and WebSearch tool to frontmatter
- Replace fictional context-manager protocol with a concrete Requirements Gathering step
- Replace vague adjective checklists with measurable criteria and named tools/thresholds
- Sharpen frontmatter description examples with concrete frameworks and a boundary-clarifying example
- Disambiguate the "prompt-engineer" cross-reference (ai-specialists vs data-ai)
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: remove stray content marker, make requirements gathering conditional
Addresses Greptile review feedback on PR #775: an unmatched </content>
marker leaked into the installed prompt, and the requirements-gathering
step re-asked for details the user had already supplied.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QS1SncnafSWXMRXQa1yVBA
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance ai-ethics-advisor based on automated review
- Fix malformed YAML frontmatter: description examples were bare unindented
text outside the description field, breaking YAML parsing / delegation
matching. Merged into a single quoted description scalar per sibling
convention.
- Add missing model: sonnet field, consistent with other ai-specialists agents
- Add NIST AI 600-1 (Generative AI Profile) subsection with the 12 GenAI risk
categories
- Add EU AI Act compliance-timeline caveat with current/pending deadlines
- Cite OWASP Top 10 for LLM Applications (LLM01, LLM06) in Agentic AI Ethics
- Add LLM/Generative System Explainability subsection (chain-of-thought
faithfulness caveat, citation-grounded transparency)
- Add Aequitas toolkit, 80% disparate-impact rule, and Bash least-privilege
clarification
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* chore: sync generated content for ai-ethics-advisor
Regenerate only the per-component content artifact for
ai-ethics-advisor so the dashboard detail view reflects the updated
frontmatter/body. Full catalog regeneration (docs/components.json,
counts, search-index, download stats) is left to the scheduled
update-json-data workflow, since a local run here pulled stale/wrong
download counts that would have clobbered production data.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018wpuLQJaC6qskfDwUSioLF
---------
Co-authored-by: Claude <noreply@anthropic.com>
Adds the PatentFig AI skill: generate patent-office-compliant figures
from text (PNG or SVG line art), vectorize drawings to SVG/DXF/vector
PDF, AI-upscale, and convert to filing-ready formats via the PatentFig
REST API. Mirrors github.com/TopLocalAI/patentfig-skill.
Claude-Session: https://claude.ai/code/session_01Eh5mM5VPK3Rqf19GnLxSBJ
Co-authored-by: davie <davie@daviechen.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* improve: enhance product-manager based on automated review
- Remove fabricated JSON "protocol" blocks and invented metrics (fake agent-to-agent JSON requests, +$4.2M/84% satisfaction canned numbers) that don't map to real Claude Code mechanics and risk priming fabricated outputs
- Add model: sonnet to frontmatter (required field for agents)
- Replace ~20 flat keyword-list sections with concrete worked frameworks: RICE scoring formula, OKR worksheet, discovery-to-launch checklist with exit criteria, one-page PRD template, and stakeholder communication cadence
- Add explicit "How This Differs From Related Agents" section disambiguating from product-strategist (market/positioning strategy) and se-product-manager-advisor (GitHub issue creation), mirrored in description examples
- Fix cross-reference from "customer-success" to the actual component name "customer-success-manager"
- Add explicit Write/Edit output guidance (PRD/roadmap markdown documents) and anti-fabrication guidance for progress reporting
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
* fix: restore component download counts reset by catalog regeneration
The prior commit regenerated the catalog without Supabase credentials,
resetting download counts to 0 for ~1800 unrelated components. Restore
each file's original download counts from main while keeping the
product-manager description update.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012oa4wnU6RLfRH4gFWLFumC
* fix: soften product-manager/product-strategist disambiguation
product-strategist's own description explicitly covers feature
prioritization and roadmap planning, so the prior wording (claiming it
handles only market/positioning docs) contradicted the referenced
agent. Acknowledge the overlap instead, per cubic review feedback.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012oa4wnU6RLfRH4gFWLFumC
---------
Co-authored-by: Claude <noreply@anthropic.com>
Claude