Skills & tools catalog

General dev workflow helpers: commit flow, skill contribution, team orchestration, image-parser and web-researcher agents, shellcheck-on-edit hook

iOS development loop: simulator build + preview + screenshot delivery, device builds, TestFlight/App Store release automation

Linear project management conventions and slash commands: issue templates, status taxonomy, agent-ready workflow, autonomous issue pickup

Turn rough prompts into deterministic specs and surface the right next step: /improve-prompt, /plan, debug/refactor/review lenses, a follow-up-suggesting Stop hook, and opt-in secret/format guardrails. Zero config; guardrails off by default.

Set-and-forget orchestrator for long Claude Code runs across multiple repos. Detects 5-hour usage limit, waits for reset, resumes every tmux session

Stage current changes, lint shell scripts with shellcheck, and create a git commit with conventional message

Contribute a change or new skill back to claude-skills seeding repo: branches, validates, commits, opens a PR

Capture a session lesson (symptom → root cause → fix → evidence) into the skills catalog: dedupes against existing skills first (extend beats duplicate), drafts a symptom-first SKILL.md, hands off to contribute for the PR

Generate a self-contained HTML explainer for the current branch/PR: what's implemented, an architecture diagram, key files, test evidence, next steps — scoped to one branch rather than the whole repo

Scan for secrets in BOTH the working tree and the full git history before flipping a GitHub repo from private to public — a diff-only scan misses secrets committed then deleted, which stay cloneable once public

Configure GitHub branch protection on a solo-owner repo so OTHER people's PRs require the owner's approval, while the owner can still self-merge — avoids the "review not required for administrators" lockout

Adding user-selectable alternate app icons to XcodeGen iOS app, keeping a new icon as default

Build iOS app on booted simulator, launch, screenshot, and deliver to user's iPhone via hybrid channel (iMessage text + iCloud Drive)

Implementing/reviewing SwiftUI Face ID/passcode app lock, avoiding four bypass/exposure pitfalls

Recording repeatable iOS feature demo videos/GIFs from simulator using paced XCUITests and simctl recordVideo

Build iOS app for simulator or connected iPhone via build.sh, encoding project's signing/profile-detection rules

Gated pre-flight (compliance strings, entitlement parity, MainActor runtime-trap audit) + Fastlane gym/pilot/deliver with xcodebuild/altool fallback: archive, validate, upload, tag, site deploy

Two Swift 6 MainActor compile diagnostics under SWIFT_DEFAULT_ACTOR_ISOLATION = MainActor: general pure-compute type migration, and the narrower synthesized Decodable/Equatable/Hashable conformance case on a zero-stored-state struct

Diagnosing opaque Xcode Cloud archive failures by validating exported .ipa locally with altool

Adding Swift Testing unit tests and XCUITest targets to XcodeGen (project.yml) iOS app

Idempotent repo standardizer: marketing copy home, Fastlane files, ci_post_clone.sh, release-hooks dir, architecture checklist, AGENTS.md skeleton. Creates what's missing, reports drift, never overwrites

The site standard (floorprint model): site/ source in the app repo → split-repo public GitHub Pages, subtree force-push over an SSH deploy key, og/CSP/favicon lint, skeleton + runbook

The four-rule ci_scripts/ci_post_clone.sh contract (materialize gitignored .xcodeproj, mirror local generation, pin Package.resolved, brew-only) plus the PR-check/tag-release workflow recipe

Fixes Teams/iMessage/Slack unfurl showing no image or wrong size, favicon missing, fonts blocked by CSP: absolute og:image + true dimensions, self-hosted fonts + strict CSP, complete favicon set

Diagnoses launch watchdog 0x8BADF00D SIGKILL + boot-loop crashes from heavy work implicitly running on MainActor at launch; off-main idioms + idempotent-retry rule

Diagnoses brk 1 crashes on SwiftUI's AsyncRenderer thread from @MainActor closures captured by UIKit (dynamic color/image providers) and invoked off-main; includes re-entrancy guard patterns

CloudKit-safe SwiftData: explicit cloudKitDatabase config, throwing container factory with fallback, single-side @Relationship inverse, avoiding NSManagedObject-reserved names, centralized schema

App→widget snapshot bridge over an App Group: one Codable DTO, atomic writes, backfill-on-launch, and the transient-empty-clobber + App-Lock double-redaction invariants that shipped as bugs before

App-Group share/action-extension handoff inbox with a per-batch attempt-cap + quarantine so one poison shared file can't boot-loop the host app

One pure resolver for all deep-link URLs; drops (not defers) links while App-Lock is engaged; validates against path traversal

Grounds on-device AI on Vision-layout text instead of PDFDocument.string (which collapses multi-column layouts and causes confabulation); versioned sidecar; verify on the cold path

Apple @Generable patterns: flat schemas only (nested hangs generation on iOS 26), verbatim-quote pinning, clipping grounding text to the model's context window

Recovers long RoomPlan/ARKit captures: persist the processed result before the hang-prone build step; harden against partial/corrupt recovery files; crash marker

Soft variance-of-Laplacian sharpness gate (not a hard reject) plus scan auto-naming discipline to stop OCR prose becoming an item name

Camera capture "gets stuck" with no error — AVCapturePhotoCaptureDelegate callback never fires after a session mid-teardown/re-mount race; add a timeout watchdog around the one-shot delegate call

A SwiftUI title placed in ToolbarItem(.topBarLeading) renders on iOS 26 as a single letter + ellipsis inside a glassy circle instead of full text — move it out of the toolbar slot or fix the Liquid Glass layout

SwiftData unit tests crash the runner, a store-reset throws "Batch delete failed due to mandatory OTO nullify inverse", or store tests race when the full suite runs together — in-memory harness fixes all three

VNDetectBarcodesRequest returns zero results for a perfectly valid QR/barcode — Vision's ML context fails to init, reliably on Simulator — fall back to CIDetector so decode tests stay green in CI

SwiftUI's TabView won't give you a labeled bottom bar AND swipe-between-tabs at once; a custom pager to get the swipe can race programmatic tab selection against deep link/NFC/Siri routing

Type-ahead suggestion chips that go stale within days on a hardcoded list, or filtering logic that can't be unit-tested and is duplicated between add/edit screens — derive vocabulary from the persistent store instead

A GitHub Pages deploy is green in CI but a linked doc page 404s live — a flat `cp docs/*.html _site/` plus top-level-only link rewriting never descends into subdirectory pages

"rsvg-convert: command not found" / no ImageMagick / no cairosvg / no Inkscape — generate favicon.ico, apple-touch-icon.png, and og-card.png from a simple geometric+glyph SVG mark with Pillow instead

Privacy/support/legal pages render with broken layout after sharing the homepage's global stylesheet (hero-only rules bleeding via bare body/html selectors), plus nav/footer links that drift page-to-page

Fan out multiple subagents to fix a batch of independent code-review findings or plan tasks in an iOS repo without them stomping the same working tree or racing xcodebuild test runs on one simulator

An app can have both an in-app What's New/changelog screen (compiled into the binary) and App Store Connect's What's New/What to Test metadata (a manual paste) — two independent surfaces that silently drift; gates the release preflight on both

Linear project management conventions and slash commands: label vocabulary, status taxonomy, branch/PR naming, issue template, .claude/linear.yml policy

Apply disciplined debugging lens: reproduce first, isolate, form hypothesis, prove with failing test

Turn rough/vague request into deterministic spec: restated goal, acceptance criteria, assumptions, recommended commands

Decompose non-trivial task into explicit tracked plan with goals and per-step acceptance criteria

Restructure code without changing behavior, guarded by tests that pass before and after

Apply code-review lens to diff/branch/file: find correctness bugs, security issues, quality problems, ranked by severity

Re-scan skills and rebuild the command-advisor registry + history profile on demand; also wires the statusline.

Orchestrate long unattended Claude Code runs across multiple repos with wind CLI, resume sessions after 5-hour usage limit

Contribute a change or new skill back to claude-skills (creates branch + PR)

Evaluate whether current task warrants a Claude Code agent team; propose composition or push back

Invoke lesson capture from any repo; reports the dedupe verdict (extended vs created) and the PR URL

Generate a self-contained HTML visual explainer for the current branch, diffed against a base branch (default main); writes to .scratch/branch-explainer/ and delivers it

Apply UI fix to iOS app (reads .claude/app.yml), then build + screenshot to prove it worked

Scaffold .claude/app.yml for this iOS app: detect scheme/bundle-id/team from project.yml or xcodebuild

Build iOS app (reads .claude/app.yml; --no-build skips), launch on booted simulator, optionally deep-link, screenshot, embed inline

Run the repo standardizer and walk drift items one by one

testflight or appstore (--dry-run supported): gated Fastlane release pipeline, stops at every failing gate

create/deploy/verify the marketing site per the standard

Add agent-blocked label to Linear issue and post comment with given reason

Bootstrap Linear PM in current repo: interactive setup of .claude/linear.yml and creation of standard label vocabulary

File new Linear issue in configured project using standard template (Why/What/Acceptance criteria/Notes)

Autonomous loop primitive: fetch issue, validate, create branch, implement work, run verify, open PR, update Linear

Show digest of current repo's Linear project: In Progress, In Review, agent-blocked, agent-ready queue, recently shipped

Reconcile current branch and open PR (if any) with Linear issue: post PR link comment, move issue to In Progress/In Review

Guide user through Second Wind setup (init → prompt → up → dash) and optionally run each step

Python CLI orchestrator for long Claude Code runs: init, prompt, add, up, status, resume, down, watch, dash, guide subcommands

Vision-capable focused image-analysis subagent: screenshots, photos, diagram analysis with OCR, comparison, text extraction

Focused research subagent: web research and library-documentation lookups via Context7 and WebFetch

PostToolUse hook: lint .sh files with shellcheck after Edit/Write, non-blocking informational output

PreToolUse hook: blocks Read/Edit/Write of secret-looking files when PROMPT_CRAFT_BLOCK_SECRETS=1

PostToolUse hook: formats just-edited file with configured formatter when PROMPT_CRAFT_FORMAT_ON_EDIT=1

After each turn, routes through the command advisor to suggest follow-up commands from git state. Surfaced to the user only (systemMessage); never fed to the model. Silent on no match.

Before each prompt, surfaces the top advisor recommendation as a user-only systemMessage. Never feeds the model.

On session start, rebuilds the registry when stale (repo change, scan-signature change, or Claude version change).

Live '💡 next: /x' statusline segment; chains to your existing statusline command. Wire with wire_statusline.py --wire.