* feat(dashboard): split component hero into info + live Claude Code terminal
Before: every component detail page opened with a full-width header card,
then the install command stacked underneath it, and nothing on the page
showed what installing the component actually does.
After: the top of the page is two columns. The left keeps the component
identity (icon, name, type, category, downloads), adds its description,
and carries the install command plus the GitHub/metadata buttons. The
right is a Claude Code terminal that types the npx install command,
prints the installer's real output, then opens `claude` and runs the
component — the skill being loaded, the agent task, the slash command,
the MCP server connecting, the hook firing on its own event, the mod
loading behind the function-hooks flag.
How: a new ComponentTerminal React island builds a per-type script from
the same lines cli-tool/src/index.js prints, and replays it with a typing
animation. It starts on scroll into view, has a Replay button, and prints
the whole transcript at once under prefers-reduced-motion. The page
passes the MCP server name / hook event and the real file count so each
demo shows the component's own handles.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JtFVg9ffrJGnvvny4sQ3rw
* fix(dashboard): match the real Claude Code startup screen in the terminal
Before: opening `claude` in the demo printed an invented "✻ Welcome to
Claude Code" panel, the prompt was a plain `>`, and the terminal had no
status bar.
After: it prints what the CLI actually shows on open — the pixel mascot
beside a bold "Claude Code" line and the working directory — types at
Claude Code's own `❯` prompt, and once the session is open the terminal
wears the CLI's status bar (directory, branch, auto mode on).
How: the welcome step renders an inline-SVG mascot and header instead of
box-drawing characters, `ask` steps use `❯`, and a `sessionOpen` flag
(set once the startup screen has printed) drives the status bar and the
trailing prompt character. The version and plan lines are left out on
purpose: they would go stale on a public page.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JtFVg9ffrJGnvvny4sQ3rw
---------
Co-authored-by: Claude <noreply@anthropic.com>
Vendors FazalAAli/pi-agent-for-claude (MIT): a pi subagent type whose
model steps run the pi CLI, so pi agents and agent-team teammates run in
Claude Code's own agent UI, streamed live, on any model pi can reach.
Component spotlight on the orchestration/move slash command (installs
as /move), covering installation, status-transition rules, and usage
examples for the todos -> in_progress -> qa -> completed workflow.
Cover image generated with Higgsfield in the existing blog cover style.
Claude-Session: https://claude.ai/code/session_01J57vKA4SLAfpA5N8tBLsmj
Co-authored-by: Claude <noreply@anthropic.com>
Syncs the vendored copy with the canonical source at context-architecture.dev
(skills/context-architecture/SKILL.md, v0.3.0). Attribution to Sergio Azócar and
the CC BY 4.0 license are unchanged.
Adds a new 'curviate' skills category (9 skills mirroring the published
curviate-plugin: dispatcher, inbox, quickstart, network, premium, profile,
jobs, engage, search) covering the Curviate CLI's LinkedIn surface, and an
MCP entry under mcps/marketing for @curviate/mcp, the hosted MCP bridge for
the same API.
Co-authored-by: Raphael Redmer <ra.redmer@outlook.com>
* improve: enhance customer-success-manager based on automated review
- Add Edit tool to frontmatter, matching Write-capable peer agents
- Add "How This Differs From Related Agents" section and reciprocal
handoff bullets with customer-support and sales-engineer
- Add illustrative composite health-score rubric plus NRR/GRR formulas
- Add churn-prediction leading indicators and CS platform/tooling prompt
- Add tiered engagement-model guidance (high-touch/pooled/tech-touch)
- Consolidate anti-fabrication rules into a dedicated section
Automated review cycle
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Nqwh2wmGQnrbEKH83wqt9x
* fix: distinguish pricing/deal ownership from legal contract review
Addresses review feedback on PR #914: the commercial (account/deal
team) and legal (legal-advisor) routing for pricing vs. contract
terms was conflated as interchangeable ("account/deal team or
legal-advisor"). Split into two explicit ownership clauses.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Nqwh2wmGQnrbEKH83wqt9x
* fix: broaden sales-engineer handoff trigger, fix section mismatch
Addresses cubic review feedback on PR #914: the handoff was gated on
"POC/RFP-won account goes live", missing deals that close without a
formal POC/RFP. Broadened to trigger on pre-sale opportunity close
(with or without POC/RFP), including onboarding/implementation
coordination, and aligned the matching description in "How This
Differs From Related Agents" to avoid the two sections disagreeing.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Nqwh2wmGQnrbEKH83wqt9x
---------
Co-authored-by: Claude <noreply@anthropic.com>
Render a "Beta" badge next to the <h1> on /mods, reusing the amber
palette of the page's existing Early Access banner. The title row
becomes a flex container so the badge sits inline and wraps on narrow
screens; other component types are unaffected.
Claude-Session: https://claude.ai/code/session_01JQ9pFGm3xwDohnhzW9ThEV
Co-authored-by: Claude <noreply@anthropic.com>
Comment out the <FeaturedSection /> render and its import in index.astro.
The component, FEATURED_ITEMS and the /featured/[slug] pages stay intact
(still linked from the sitemap), so re-enabling is a two-line uncomment.
Claude-Session: https://claude.ai/code/session_01JQ9pFGm3xwDohnhzW9ThEV
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance competitive-analyst based on automated review
- Add a Report Structure section defining the default deliverable format
(executive summary, per-competitor profile, benchmarking matrix, SWOT,
opportunities/threats, recommendations, sources appendix) plus a
condensed battlecard option for sales-facing requests
- Add a monitoring-cadence practice recommending quarterly refresh plus
event-triggered re-analysis (M&A, funding, leadership, launches)
- Extend intelligence-gathering practice to include internal win/loss
notes and CRM exports as a sourced input
- Add reciprocal scope-boundary guidance deferring to market-researcher
and trend-analyst for market sizing and cross-industry analysis
- Normalize description phrasing to match catalog convention
Automated review cycle
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018TfEsDWqgpD3TuUgWNq899
* fix: add confidentiality safeguard for internal win/loss data
Addresses greptile review finding on PR #908: the new win/loss/CRM
intelligence-gathering guidance had no rule against surfacing
confidential customer/deal details verbatim in deliverables.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018TfEsDWqgpD3TuUgWNq899
* fix: consolidate confidentiality guidance for internal CRM data
Integrates greptile/cubic review feedback: the intelligence-gathering
instruction now explicitly frames win/loss/CRM data as authorized
internal context and defers to the single confidentiality rule in
Ethical & Legal Boundaries, removing a redundant/mismatched restatement.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018TfEsDWqgpD3TuUgWNq899
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance web3-integration-specialist based on automated review
- Update default stack guidance from wagmi v2 to wagmi v3 (viem stays v2.x)
- Add EIP-5792 batch-transaction support (wallet_sendCalls/useSendCalls)
- Add ERC-7715 (wallet_grantPermissions) as the standardized session-permission mechanism
- Name concrete decentralized storage tooling (Pinata, Storacha, Filebase, Arweave/Filecoin)
- Reorder wallet-library recommendation to lead with Reown AppKit given RainbowKit's uncertain wagmi v3 support
Automated review cycle
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L5gRKyfuGhjveRFoxg9xvp
* fix: address review feedback on batch failure reporting and Filecoin persistence
- Preserve per-call failure details in EIP-5792 batch state guidance,
since batches are not necessarily atomic
- Qualify Filecoin as requiring an explicit renewal/replication strategy
for guaranteed persistence, rather than treating it as inherently permanent
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L5gRKyfuGhjveRFoxg9xvp
---------
Co-authored-by: Claude <noreply@anthropic.com>
The skill's "code the user will execute" exemption was blanket: it
suppressed the discernment nudge for any code the user would run,
including scripts that delete data, expose secrets, or change
production state, where merely running the code doesn't verify those
risks.
Restrict the exemption to trivial, local, reversible code, and keep
nudging code that affects production, external systems, security, or
data.
Addresses a cubic-dev-ai review finding (P1, confidence 8/10) on PR
#898 that landed after the PR was merged, so following up here rather
than reopening the merged PR.
Claude-Session: https://claude.ai/code/session_012T1LgFb24v87ZAUk9RjP6C
Co-authored-by: Claude <noreply@anthropic.com>
- Add generated-files-guard.yml: fails non-maintainer PRs that commit
docs/components.json or dashboard/public catalog artifacts and posts
revert instructions (pull_request_target, API-only, no PR code checkout)
- Welcome bot now warns up front not to commit generated files
- CLAUDE.md: document who regenerates the catalog (maintainers only) and
why contributor PRs must not include it
Claude-Session: https://claude.ai/code/session_01JVP5nv7a7bMkwvy67LgtPF
* feat: Sync new skills from anthropics/skills
Migrated from github.com/anthropics/skills (Apache 2.0):
- Add `academy-guide` (productivity) - recommends matching Claude Academy
courses/tutorials for "how do I use Claude" style questions
- Add `discernment-nudge` (productivity) - prompts 2-3 follow-up questions
after substantive answers to help users check facts/reasoning before acting
- Refresh `frontend-design` (creative-design) SKILL.md with the newer,
more detailed upstream guidance on aesthetic direction and self-critique
- Fix placeholder copyright line in `web-artifacts-builder` LICENSE.txt to
match upstream ("Copyright 2026 Anthropic, PBC.")
All other skills already present locally were compared against upstream
and left untouched: several (docx, pptx, xlsx, mcp-builder, skill-creator,
claude-api) have local customizations beyond the upstream reference
versions, and the rest (algorithmic-art, brand-guidelines, canvas-design,
internal-comms, theme-factory, slack-gif-creator, webapp-testing,
doc-coauthoring, web-artifacts-builder) are unchanged from upstream.
Regenerated the component catalog (docs/components.json,
dashboard/public/components.json and split artifacts) via
scripts/generate_components_json.py, which also refreshed live download
counts from Supabase.
Validated with the component-reviewer agent: no critical issues.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012T1LgFb24v87ZAUk9RjP6C
* chore: regenerate catalog after merging main
Claude-Session: https://claude.ai/code/session_01JVP5nv7a7bMkwvy67LgtPF
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance smart-contract-auditor based on automated review
- Disambiguate description/role from smart-contract-specialist (architecture) and blockchain-developer (implementation)
- Add oracle manipulation and proxy/storage-collision vulnerability classes to Focus Areas
- Add standard-specific vuln classes: ERC-4626 inflation, ERC-2612 permit replay, ERC-4337 UserOp/paymaster bypasses, ERC-777/1363 callback reentrancy
- Name read-only reentrancy and EIP-7702 delegation risks explicitly under attack pattern recognition
- Add anti-hallucination "report only observed findings" delivery summary
- Add Non-EVM Audit Notes section scoping the EVM/Solidity toolchain and pointing to platform-native equivalents
- Add exploit-tracker/audit-contest cross-referencing step to the Approach
Automated review cycle
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KoYeVdPeQTArVRZH1WpDkr
* fix: address cubic review feedback on smart-contract-auditor
- Reframe the ERC-2612 Focus Areas bullet so it no longer implies
ordinary permit front-running is itself a vulnerability — EIP-2612's
domain separator/chainId and nonce already prevent replay; scope the
guidance to permit correctness and allowance inheritance instead
- Move the "Delivery summary" anti-hallucination clause out of the
Non-EVM Audit Notes subsection into its own Delivery Summary heading
so it clearly applies to every audit, not just non-EVM ones
Addresses review feedback from cubic-dev-ai on PR #897.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KoYeVdPeQTArVRZH1WpDkr
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance content-marketer based on automated review
- Remove three broken agent references (social-media-manager, pr-manager,
brand-manager) and replace with real agents (social-media-copywriter,
business-analyst/data-analyst)
- Remove fabricated-metrics example content in progress tracking and
delivery notification; replace with placeholders requiring real,
measured data
- Replace the legacy context-manager JSON handshake with a concrete
"When Invoked" checklist that asks the user directly for brand voice,
audience, metrics, and goals
- Add "Use this agent PROACTIVELY" framing to the description
- Add differentiation notes vs. seo-specialist, marketing-attribution-analyst,
and social-media-copywriter in example commentary
- Add AI-referral-traffic tracking as a baseline analytics practice
- Reword E-E-A-T block to emphasize content-marketer's drafting role vs.
seo-specialist's audit role
Automated review cycle 108 | session review-20260913-130642
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QhAPA3ELVf1Z2g4N3dutdc
* fix: remove mismatched social-media-copywriter handoff, tighten intake wording
social-media-copywriter is hardcoded to promote episodes of a specific
podcast (podcast-creator-team/social-media-copywriter.md), so it was not
a suitable general handoff target for content-marketer. Also tightens
the "When Invoked" step 1 to only ask for missing/ambiguous details
instead of unconditionally re-asking for everything.
Addresses review feedback from greptile-apps on PR #896.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QhAPA3ELVf1Z2g4N3dutdc
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance mlops-engineer based on automated review
- Replace fabricated metrics in progress tracking/delivery notification with placeholders
- Rewrite checklist to require negotiated/verified targets instead of fixed asserted numbers
- Add concrete Tooling ecosystem section (MLflow, Kubeflow, Feast, KServe, KEDA, Argo CD, etc.)
- Add hand-off/boundary language vs ml-engineer, machine-learning-engineer, ai-engineer
- Add upfront requirements-gathering checklist before proposing architecture
- Add model: sonnet and WebSearch tool to frontmatter
- Name concrete tools in Security for ML section (Trivy, Vault, OPA/Gatekeeper)
- Fix stray double blank line after opening persona paragraph
Automated review cycle | Co-Authored-By: Claude Code <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018zsvYhdEgX7VGvHYEwMNQn
* fix: align description examples with negotiated-target semantics
cubic review flagged that the frontmatter description's example
outcomes still asserted the exact fixed numbers ("30 minutes",
"above 70%") that the new checklist explicitly says not to assert.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018zsvYhdEgX7VGvHYEwMNQn
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance business-analyst based on automated review
- Add proportional/right-sizing guidance for BRD and process diagrams
based on initiative size and stakes
- Add a User Story Template (Given/When/Then acceptance criteria)
alongside the existing BRD template for smaller/agile requirements
- Add explicit tool-usage guidance for WebSearch/WebFetch/Glob/Grep
- Version-pin the BABOK reference (v3 + Agile Extension) and note
agile BA practice (backlog refinement, epic/story writing)
- Regenerate component catalog
Automated review cycle
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H22Gr3YjoDwqEdC6XQ2Boh
* fix: add priority/owner/traceability fields to user story template
The Process Modeling Approach section requires every requirement to
carry a MoSCoW priority, a named owner, and a traceability link, but
the new User Story Template omitted all three even though it's offered
as a lighter alternative to the BRD for the same kind of requirements.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H22Gr3YjoDwqEdC6XQ2Boh
* fix: drop unrelated catalog regeneration noise from PR
The full catalog regen picked up download-count churn across every
component type, unrelated to the business-analyst doc change. Revert
those files to main and let the dedicated catalog-generation workflow
update them separately.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H22Gr3YjoDwqEdC6XQ2Boh
---------
Co-authored-by: Claude <noreply@anthropic.com>
* improve: enhance hackathon-ai-strategist based on automated review
- Add explicit Write tool usage guidance (when to persist artifacts vs. stay conversational)
- Move demo narration drafting to Phase 1 so builds stay anchored to what will be shown; Phase 5 now refines rather than originates it
- Add the 80/20 delegation heuristic for AI-assisted rapid-prototyping tools
- Include rapid-prototyping platforms in the "verify currency via WebSearch/WebFetch" instruction
Automated review cycle
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01681WTwqbqv5emwagAftBw5
* fix: derive Phase 1 demo narration length from the confirmed submission cap
Addresses a greptile review finding on PR #876: the Phase 1 narration
draft was hardcoded to ~90 seconds regardless of the actual submission
cap gathered in the Required Initial Step, conflicting with Phase 5's
existing instruction to time against the real cap rather than an
assumed 3 minutes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01681WTwqbqv5emwagAftBw5
---------
Co-authored-by: Claude <noreply@anthropic.com>
The install section documented `--function-hook` as available, but the flag
was only ever merged to main - it is not in the published 1.29.4 on npm, so
`npx claude-code-templates@latest --function-hook ...` failed with
"unknown option". Bump to 1.29.5 so the flag ships, state the minimum
version in the article, and add the manual plugin layout as a fallback.
Also correct claims that testing on Claude Code 2.1.266 contradicted or
that the binary now answers outright:
- The runtime is compiled in. Document the real $ surface found in the
binary (fs/http/process/model/store/session/ui/tool/agent/command/prompt),
the classic.* event namespace and engine.create withholding, replacing
the "what if the flag does nothing" box.
- Soften the <name>@skills-dir auto-load claim; it is unverified. Recommend
--plugin-dir, which is the path actually tested.
- Note secret-redactor's two real limits: shape-only matching (PASSWORD=
hunter2 passes) and a connection-string pattern that stops at the @, so
host, port and database name survive redaction.
- Note that $.ui.log appears TUI-only and is silent under `claude -p`;
universal-audit-log is unaffected since it writes via $.fs.append.
- Hang budget: the binary has reentrancy detection and per-plugin model
budgets, so answer the part that is now known and keep the rest open.
Claude-Session: https://claude.ai/code/session_01KnCWDj7sV9JznrV5UMNy8W
* improve: enhance architecture-modernizer agent based on automated review
- Add required `model: sonnet` field (was missing, would fail component review)
- Add `Glob` to tools list for consistency with sibling agents and to
support file/module enumeration during service-boundary analysis
- Add closing "Integration with Other Agents" hand-off section reciprocating
legacy-modernizer's existing reference to this agent
- Add concrete tool-usage guidance (Grep/Glob for module/coupling
inventory, Bash for dependency-graph tooling, Write for ADRs/diagrams)
- Strengthen the closing risk/rollback guidance with named techniques
(contract tests, shadow traffic, feature flags, canary rollout, legacy
decommissioning, shared-database coupling)
- Add tactical DDD context-mapping detail to the Approach section
- Tie Output bullets to concrete artifact formats (ADRs, Mermaid, OpenAPI/AsyncAPI)
Automated review cycle
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KRgMrRbDvzACuMDfhSHKnE
* fix: make ADR output location adapt to target repo convention
Address Greptile review feedback on PR #874: hardcoding docs/architecture/
ignores a target repo's existing ADR convention (e.g. docs/adr/,
docs/decisions/), creating a second decision-record hierarchy.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KRgMrRbDvzACuMDfhSHKnE
---------
Co-authored-by: Claude <noreply@anthropic.com>
claude[bot]