Wednesday Agent Skills

AI skills for Wednesday Solutions projects — git discipline, PR automation, terminal dashboard, greenfield planning, and brownfield codebase intelligence with real-time chat, drift detection, and test generation.

1. Installation

Requirements

• Node.js ≥ 18
• npm ≥ 8

Option 1 — npx (no setup)

npx @wednesday-solutions-eng/ai-agent-skills install

Option 2 — global

npm install -g @wednesday-solutions-eng/ai-agent-skills
wednesday-skills install

Option 3 — shell (no npm)

bash install.sh

Run in your project root. The installer:

1. Copies skills into .wednesday/skills/
2. Writes agent config files (CLAUDE.md, GEMINI.md, .cursorrules, .github/copilot-instructions.md)
3. Installs git hooks (post-commit, post-merge) for automatic graph updates
4. Prompts for optional coverage and Sonar integration
5. Symlinks skills into ~/.claude/skills/ for Claude Code discovery

2. Configuration & AI Tools

No API key needed to use skills inside Claude Code, Cursor, or Gemini CLI. When inside an AI IDE, the IDE acts as the intelligence engine — skills are standard instructions, not local scripts.

Supported AI tools

| Tool | Configured via | |------|---------------| | Claude Code | CLAUDE.md | | Gemini CLI | GEMINI.md | | Antigravity | ~/.gemini/antigravity/skills/ (run wednesday-skills sync) | | Cursor | .cursorrules | | GitHub Copilot | .github/copilot-instructions.md |

Environment Variables

API keys are only required for standalone CLI workflows (plan, summarize, gen-tests). Run the interactive configuration wizard:

wednesday-skills config

Or manually add to .env:

• OPENROUTER_API_KEY or ANTHROPIC_API_KEY: Used by offline LLM-backed tools.
• GITHUB_TOKEN: Used by wednesday-skills dashboard to fetch PR data.

3. Every Skill Explained

Core Workflow Skills

| Skill | What it does | |---------|-------------| | git-os | Enforces conventional commits — no bad or ambiguous commit messages allowed. | | pr-review | Gemini fix queue — categorizes PR comments by impact, applies fixes upon dev approval. | | deploy-checklist | Walks through pre-deploy checks and post-deploy monitoring checklists. | | wednesday-dev | Enforces import ordering, file complexity limits (max 8), and naming conventions. | | wednesday-design | Asserts the use of 492+ approved UI components, design tokens, and animation patterns. | | sprint | Translates ticket IDs into git branches, PR titles, and description templates automatically. | | greenfield | Parallel AI personas (Architect, PM, Security) produce a comprehensive PLAN.md in minutes. |

Brownfield Intelligence Skills

| Skill | What it does | |---------|-------------| | brownfield-chat | Plain-English codebase Q&A using structural graphs (zero hallucinated data). | | brownfield-query | Deterministic structural queries returning dependencies, endpoints, and file metrics from SQLite (graph.db). | | brownfield-fix | Calculates Risk score + blast radius before the AI is allowed to edit a file. | | brownfield-drift | Enforces architecture boundaries defined in PLAN.md preventing domain spillage. | | brownfield-gaps | Enhances dynamic runtime graph coverage via localized subagents. |

Feature Deep Dives

🧠 Terminal Dashboard (wednesday-skills dashboard)

Provides an interactive CLI interface for tracking open PRs, unassigned semantic fix queues, installed skills status, and detailed LLM token cost breakdowns.

💰 Token Cost & Savings Report

Every LLM-backed command (map, summarize, gen-tests, etc.) automatically prints a cost report after it runs:

━━━ Token Usage Report ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Command:       map
  LLM calls:     18   (6 cache hits → 0 tokens)
  Tokens used:   9,240  (in: 6,800 / out: 2,440)
  Baseline est:  54,000  (cost of reading raw files)
  ▼ 44,760 tokens saved  (82%)
  Cost:          $0.0013  (baseline: $0.1620 vs Claude Sonnet)
  ▼ $0.1607 saved by using this model
  ──────────────────────────────────────────────────
  Operation          Used  Baseline   Saved%  Calls
  arch-overview     1,320     6,000    78%  1
  summarize         5,480    15,000    63%  12 +6cached
  gap-fill          2,440     9,000    72%  5
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Baseline is computed as what Claude Sonnet ($3/M tokens) would spend reading the equivalent raw source files directly. Actual cost reflects the real model used (e.g. gemini-2.5-flash-lite at $0.10/M). The difference is your savings from both the pre-computed graph and the cheaper model selection.

→ Full details: docs/token-cost-report.md

🛡 Architecture Drift Detection (wednesday-skills drift)

Validates that the actual code structure safely adheres to the constraints in PLAN.md (e.g. frontend-never-imports-db, no-circular-deps). Designed to plug into CI/CD pipelines to block architectural decay during PRs.

🧪 AI Test Generation (wednesday-skills gen-tests)

Generates comprehensive tests using actual callers and real AST mock behavior to safely wrap high-risk, un-covered files, using historical bug-fix commits as context.

💬 Codebase Chat (wednesday-skills chat "your question")

Ask plain-English questions ("Who wrote the auth layer?", "What breaks if I rename X?") and receive verified answers parsed instantly from local ASTs and Git history. Saves 100% of LLM tokens by using offline parsing.

4. Scenario-Based Use Cases

All workflows run entirely inside Claude Code or Gemini CLI. The IDE loads the relevant skills seamlessly and intuitively based on your conversation.

A. The Greenfield Project (Starting from scratch)

Scenario: You have an idea and need an architectural robust plan.

1. You say: "Plan this project: Build a realtime collaborative text editor."
2. AI Action: Loads the greenfield skill. The framework spins up parallel Architect, PM, and Security AI personas.
3. Outcome: A comprehensive .wednesday/plans/PLAN.md with system architecture, security risks, phased tickets, and architectural boundaries.
4. Follow Up: "Start the first ticket." → Loads the sprint skill → Sets up your git branch and PR draft automatically.

B. First Time on an Existing Codebase (Brownfield Intelligence)

Scenario: You just inherited a completely undocumented, legacy codebase.

1. You say: "Map this codebase completely."
2. AI Action: Automatically triggers wednesday-skills map --full, parsing thousands of files dynamically into a high-performance .wednesday/graph.db database.
3. Outcome: Generates MASTER.md containing global architectural user flows, and summaries.json for natural language querying. Future operations no longer require passing thousands of tokens of context files; the agent directly queries the pre-computed SQLite graph.

C. New Developer Onboarding

Scenario: A new developer joins the backend API squad and is confused.

1. You say: "Generate an onboarding guide for the backend."
2. AI Action: Resolves the onboard intent, and utilizes the SQLite Recursive CTE framework to trace deeply nested request flows natively from CLI entry points to the core domain logic.
3. Outcome: Provides an extremely focused, functional step-by-step Mermaid diagram execution flow and reading guide specifically for the exact layer the developer needs to touch.

D. Fixing a Bug in a High-Risk Monolith

Scenario: You've been tasked to fix a bug deep within a massively coupled module.

1. You say: "Fix the token expiration bug in auth.ts."
2. AI Action: Pre-emptively invokes brownfield-fix to calculate the blast radius of auth.ts inside graph.db before writing any code.
3. Outcome: If the file has a critical impact score (>80), Claude forcefully pauses, refuses to touch the code, and dictates the cascading components explicitly, asking for developer verification before proceeding safely. Once approved, handles commits using the pristine git-os skill.

E. Automating PR Code Reviews and Fixes

Scenario: The lead dev left 5 semantic code review comments on your Pull Request.

1. You say: "@agent fix #2 and #4" or "@agent fix all"
2. AI Action: Loads pr-review, parses the exact GitHub comments, structures security, safety, and style impacts, and isolates each fix into discrete, clean git commits.
3. Outcome: PR feedbacks are iteratively satisfied without polluting commit history or breaking CI checks.

F. Improving Code Coverage Programmatically

Scenario: A mission-critical module handles real-time payments but has 0% test coverage.

1. You say: "Generate tests for uncovered critical files."
2. AI Action: Evaluates gen-tests --min-risk 75 to rank code that is both completely uncovered and possesses a terrifying blast radius.
3. Outcome: Produces functional, mocked .test.js files perfectly integrated into your framework, using deterministic AST connections instead of hallucinatory scaffolding.

5. CLI Commands Reference

# Setup
wednesday-skills install                  # install + configure all agents
wednesday-skills config                   # interactive API key and model setup
wednesday-skills sync                     # re-sync all tool adapters

# Intelligence
wednesday-skills map --full               # Complete AST extraction and flow inference into graph.db
wednesday-skills onboard                  # Contextual, step-by-step interactive flows
wednesday-skills drift                    # Validates architecture against PLAN.md
wednesday-skills drift --since HEAD~5     # Run drift checks on a specific diff (PR verification)
wednesday-skills chat "question"          # Instantly ask codebase questions using BFS limits

# Analytics
wednesd