Gordon OS — Skills Library v0.3

Composable AI skills for Claude Code. Director-level portfolio management, client intelligence, agent fleet management.

Install

New machine (from npm — recommended):

npx @gordon.chan.tavant/claude-skills install --global

Installs all skills to ~/.claude/skills/, sets skipDangerousModePermissionPrompt in ~/.claude/settings.json, and copies AWS infra scripts to ~/.claude-agent/.

From source (if you have the workspace):

cd ~/Workspace/.claude/skills
npx . install --global

Other install options:

npx @gordon.chan.tavant/claude-skills install --global --update     # re-install / update all
npx @gordon.chan.tavant/claude-skills install --global speckit-specify  # single skill
npx @gordon.chan.tavant/claude-skills install --global --layer speckit  # one layer only
npx @gordon.chan.tavant/claude-skills list                          # browse all skills

AWS infrastructure (ECS, EC2 desk, Aurora, S3 sync) — run once after install:

aws configure sso --profile <your-profile>   # first-time AWS setup
aws sso login --profile <your-profile>       # login
cd ~/.claude-agent
AWS_PROFILE=<your-profile> S3_BUCKET=<your-bucket> ./setup-new-user.sh

For full AWS infrastructure setup (ECS agents, EC2 desk, Aurora, S3 sync), see resources/aws-agent/README.md.

Architecture

Skills are executable prompt instructions loaded by Claude Code. No code runs — Claude is the runtime.

Gordon types: /g-deep-research Anchor Loans
                      ↓
        loads g-research(depth=full)        → findings markdown
        loads g-extract(mode=pain-points)   → pain-point mapping
        loads g-write(type=intel)           → intel file (quality gate built in)
        loads g-extract(mode=entities)      → entity list
                      ↓
        asks: "Fan out to 4 entities? (P0: Kiavi, Pretium)"

Three layers:

• Core — single-verb building blocks with mode flags. Slash-invocable or chainable.
• Compound — pipelines that chain core skills. Markdown flows step to step.
• Utility — standalone tools outside the g-* system.

Core Skills (5)

| Skill | Modes | What It Does | |---|---|---| | /g-research | full quick refresh | Web research engine — 37 queries across 9 dimensions (industry, people, financials, competitors, pain points, news, tech stack, org, culture) | | /g-extract | signals entities patterns pain-points | Pull structured intelligence from any text. Delegates entity enrichment to g-lookup. | | /g-lookup | connections patterns conflicts playbook-match | Cross-reference signals against PORTFOLIO.md, intel files, playbooks. Surfaces what's related and what playbook applies. | | /g-write | daily intel playbook deep-dive presales | Write any canonical output file with built-in quality gate. Fails loudly if sections are missing or filler. | | /g-portfolio | read score patch read-and-score | PORTFOLIO.md state management — read current state, score engagement health, apply targeted patches. |

Compound Skills (7)

/g-morning-brief

Full morning intelligence briefing. Runs autonomously — reads Outlook, Calendar, and Teams via Chrome DevTools MCP.

open-browser → read-inbox → read-calendar → read-teams
  → g-extract(signals) → g-lookup(connections)
  → g-write(daily) → g-portfolio(patch)
  → [optional: deep-dive triggers]

Output: research/daily/daily_intel_YYYY-MM-DD.md + PORTFOLIO.md patches

/g-deep-research <company>

Full client intelligence cycle from scratch.

g-research(full) → insider-check (asks Gordon)
  → g-extract(pain-points) → g-write(intel)
  → g-extract(entities) → [g-fan-out-research]

Output: research/intel/{Company}_Client_Intel_{Month}_{Year}.md Minimum bar: 5 unique sources, 3 evidenced pain points, named contacts.

/g-presales-pack <client>

Assemble a client-tailored presales artifact for a meeting or proposal.

read-client-intel → g-lookup(playbook-match)
  → g-extract(pain-points) → g-write(presales)

/g-weekly-review

Monday AM portfolio health check.

read-daily(last 5) → g-portfolio(read-and-score) → g-portfolio(patch)

/g-flywheel-extract

Extract recurring cross-client patterns and codify as reusable playbooks.

read-intel(all) → g-extract(patterns) → g-write(playbook)

/g-fan-out-research

Scan an intel file for named entities, spawn parallel research for each.

g-extract(entities) → prioritize (P0/P1/P2/P3)
  → g-research(per entity) → g-write(intel, per entity) → consolidate

/distill-sales

Distill sales pattern intelligence from Notion meeting transcripts.

scan-notion → read-daily → g-extract(signals) → refine → write-updates

Utility Skills (4)

/agents

Universal agent interface. The only command you need for the full fleet.

/agents                               Dashboard (ECS tasks + desk chats + recent history)
/agents "research Anchor Loans"       Dispatch work (auto-routes ECS vs desk)
/agents chat anchor-loans "continue"  Named persistent desk session
/agents list                          What's running right now
/agents logs <name> [--follow]        Tail logs (ECS: CloudWatch, desk: tmux capture)
/agents stop <name>                   Stop a task or chat
/agents history [7d|30d|all]          Session history from Aurora
/agents analytics                     Usage breakdown by type and day

Routing is invisible: quick/scheduled tasks → ECS Fargate; long/iterative/research → EC2 desk (tmux window). Detects existing desk chats on the same topic and prompts before starting fresh (Enter = use existing).

/task

Fire an ephemeral ECS Fargate agent directly.

/task run /g-morning-brief
/task run "summarize all intel and update PORTFOLIO.md"
/task list
/task status

Superseded by /agents for most use — /agents auto-routes.

/safe-commit

Pre-commit safety scan + smart commit. Scans for secrets, updates .gitignore, stages, commits with conventional message, smart-tags.

/pptx-screenshot

Convert every slide of a PPTX to PNG images for visual inspection or documentation.

/pptx-screenshot /path/to/deck.pptx
/pptx-screenshot /path/to/deck.pptx slides=1-5

Uses LibreOffice if installed (pixel-perfect), falls back to python-pptx.

Speckit Skills (9 + 5 git)

Structured software specification and implementation workflow.

| Skill | What It Does | |---|---| | /speckit-clarify | Interactive clarification before specification | | /speckit-specify | Create/update feature specification from natural language | | /speckit-plan | Generate design artifacts and implementation plan | | /speckit-tasks | Generate dependency-ordered tasks.md from design artifacts | | /speckit-implement | Execute implementation plan task by task | | /speckit-analyze | Cross-artifact consistency and quality analysis | | /speckit-checklist | Generate implementation checklists from tasks.md | | /speckit-constitution | Define and enforce project coding standards | | /speckit-taskstoissues | Convert tasks.md entries to GitHub issues | | /speckit-git-commit | Structured commit with conventional commits | | /speckit-git-feature | Create and manage feature branches | | /speckit-git-initialize | Initialize new repo with standard structure | | /speckit-git-remote | Configure and manage git remotes | | /speckit-git-validate | Validate git state before merge or deploy |

Skills Sync to Remote Agents

When skills are updated locally, they reach remote agents automatically:

| Location | Syncs to S3 | Restored on remote | |---|---|---| | ~/Workspace/.claude/skills/ | rclone bisync every 5 min | ECS mount at startup, desk boot | | ~/.claude/skills/ | launchd daemon every 5 min | ECS startup, desk boot |

After installing a new marketplace skill (~/.claude/skills/new-skill/): available on the next ECS task run and next desk boot with no manual step.

Force-push immediately:

rclone sync ~/.claude/skills s3-workspace:gordonchan-claude-workspace/.claude-global/skills --progress

Design Principles

1. Verb-first, not layer-first — 5 powerful skills beat 15 thin ones.
2. Serial pipeline, markdown handoff — each step loads one skill, passes markdown output forward. No shared state.
3. Claude is the runtime — skills are prompt instructions, not code.
4. Quality built in — g-write validates every output. No separate gate needed.
5. Self-healing — /agents detects missing infrastructure and surfaces the exact inline fix.

Workspace Layout

Skills read from and write to these locations by convention:

~/Workspace/
├── PORTFOLIO.md              Active engagement tracker
├── CLAUDE.md                 Project instructions
├── project/                  Client project directories
├── research/
│   ├── daily/                daily_intel_YYYY-MM-DD.md
│   └── intel/                {Company}_Client_Intel_*.md
└── org/
    └── playbooks/            Reusable frameworks

Version History

| Version | Change | |---|---| | 0.1.0 | Initial skill library (16 skills) | | 0.2.0 | Added primitives layer, standardized contracts (26 skills) | | 0.3.0 | Merged by verb. Added agents, task, desk, pptx-screenshot. npx installer with --global and init. |