karlorz-agent-skills

Source-triaged research engine with local-first evidence, grok-search freshness checks, Context7/API verification, tiered model pinning (sonnet/haiku agents for external research, parent model for synthesis), deep-fetch, content refinement, and flexible output: terminal, file, or skillwiki vault integration with schema-compatible follow-up queueing. Spawnable as a subagent (model: sonnet) via deep-research:deep-research. v2.4.2: latest-by-default source triage, grok-search freshness, and Freshness & Verification Status reporting.

Code review for reuse, simplification, efficiency, and altitude

Generic single-pass PRD + skillwiki dev cycle engine. v1.26.16: hide dev-loop companion helpers from user command surfaces; standardize /dev-loop and $dev-loop mode entrypoints. v1.26.15: sdd-execute-worker adapter for superpowers:subagent-driven-development EXECUTE step. v1.26.14: /dev-loop dashboard mode dispatch. v1.26.9: why-skipped CLI. v1.26.8: read-only /dev-loop config-lint mode with dev-loop-config-lint.v1 reports. v1.26.7: read-only /dev-loop status mode with dev-loop-status.v1 JSON and markdown reports.v1.26.6: stable release for hardened preflight inventory signals and implemented-evidence false-positive filtering. v1.26.5: stable release for host-aware office-hours repo resolution, vault-only --all-projects discovery, and explicit degraded repo evidence states. v1.26.4: stable release for office-hours cross-project discovery and --all-projects inventory. v1.26.3: clarify investigate queue schema probes as disposable and make raw-capture fallback explicit for current SkillWiki schemas. v1.26.2: stable release for stale implemented capture hygiene and office-hours recheck. v1.26.1: office-hours Codex question flow uses request_user_input only in Plan mode and numbered fallback in Default mode. v1.26.0: compact office-hours prompt with first-run anchoring fix. v1.25.0: attended office-hours companion skill for requirements intake before prep, promotion, or /goal; sync-plugin-cache.sh and release-tooling tests updated. v1.25.0: attended office-hours companion skill for requirements intake before prep or /goal. v1.24.9: Codex dispatch correction release ships task_name-based spawn_agent docs/tests, instruction-level PLATFORM_DISPATCH wording, and Codex custom-agent TOML caveats. v1.24.8: PLATFORM_DISPATCH documents an instruction-level dispatch routing contract (Agent vs spawn_agent vs inline) at REFRESH so Codex subagent dispatch follows the live tool surface; codex-tools.md troubleshooting section corrected. v1.24.7: simplify-worker reads and follows simplify:simplify as source of truth, with structural release-tooling checks for the review-gate contract. v1.24.6: required simplify:simplify review gate prefers the simplify-worker subagent adapter with inline fallback only when worker dispatch is unavailable. v1.24.5: portable config cleanup, doctor-worker spawn fallback, and skill-relative Codex reference packaging. v1.24.4: portable SkillWiki vault auto-resolution via `vault: auto`, `skillwiki path` precedence, validated `~/wiki` fallback, and explicit-path mismatch warnings. v1.24.3: fix .claude-plugin skills path (./skills/ not ./) so setup-dev-loop/investigate/research show as user-invocable in CLI. v1.24.2: all-active-plugin version sync and companion skill schema coverage. v1.24.1: simple CI/main-first setup contract. v1.24.0: preflight prep mode inventories active work, batches human readiness questions, writes automation-ready metadata, and never starts /goal. v1.23.1: fixes Codex manifest version bumping, dependencies.yaml cache sync, and doctor-worker dependency probes for plugin-root skills/frontmatter-named agents. v1.23.0: Codex CLI/App compatibility — references/codex-tools.md maps worker Agent spawns to spawn_agent/wait_agent, documents the multi_agent config gate, and the detached-HEAD Codex App sandbox-finishing contract; .codex-plugin manifest brought into the version-bump set. v1.22.2: investigate/deep-research queue contracts are schema-adaptive — use proposed work items only when validation supports them, otherwise raw transcript captures. v1.22.0: /goal compatible mode — ORCHESTRATION_CAPS for goal-context detection, GRILL goal_override:never suppresses interviews under /goal, RETRO emits continuation hints, /goal Integration documents grill-me → /goal → dev-loop. v1.21.0: investigate mode adds proactive finding creation (`/dev-loop investigate`) from code/vault/transcript/deep-research sources with tiered queued output and vault-required gating. v1.20.0: compact-counter visibility — proactive 0-count emit, tuned thresholds {0-1 silent emit, 2 note, 3 warn, 4+ block}, HUD bridge writes ~/.claude/dev-loop/last-doctor.json for ccstatusline consumers. v1.19.0: release_policy.trigger_globs consumer in PUSH step — opt-in auto-bump on shippable commits, skip on vault/doc-only cycles. v1.18.1: vault-local wiki-presync skill discovery — probe and invoke before vault push. v1.18.0: peer-aware vault push gate — SAVE/MERGE acquire skillwiki advisory lockfile. v1.17.1: doctor-worker skillwiki bridge + research-worker stale buckets + wiki-sync dependency. v1.17.0: vault auto-commit gate + AUDIT dirty-tree check. v1.16.0: auto-archive originating raw transcripts via `closes:` frontmatter. v1.15.0: pluggable multi-backend code review. MERGE CI gate uses ci-health-worker. Execution subagents hard-wired to sonnet. /dev-loop setup scaffolds 14 config sections (A-N).

Guide for using agent-plugin-porter to convert Claude Code plugins into Codex and Agy/Gemini packages, verify generated output, install Codex personal-marketplace packages, and prepare npm publish/release workflow checks.

Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.

Hermes Agent CLI commands reference — full terminal command surface for hermes-agent by NousResearch

Browser automation via playwright-cli with browser-worker agent (model: sonnet) for mechanical Chrome lifecycle and interaction tasks — fork of microsoft/playwright-cli with attach+headed defaults and project config overrides

Host-level backup and restore with profile system (presets + custom YAML profiles), model-aware agents (sonnet worker), post-discovery research, and skillwiki infrastructure capture. v3.6.3: explicit SSH/Tailscale identity restore for post-reinstall hosts, with safety snapshots, ca-certificates handling, and SSH validation. Supports Caddy domains, databases, systemd services, Hermes snapshots, and restore target pre-inspection.

Relentless plan/design interview skills adapted from Matt Pocock's grilling, with inline cross-platform question style. v1.0.0: packages both skill names for direct invocation.