The Agents

Purpose-built AI agents for every phase. Each agent has a specific role, defined capabilities, and clear outputs.

Specialist Agents

Called in when a feature needs deep expertise in a specific area.

Claude has evolved fast since this toolkit was first built — code review, security review, simplification, and context management are now native to Claude Code. v9.0 re-worked the catalog around that: every agent below owns territory the platform doesn't cover, and the ones it absorbed are retired with honor at the bottom of this page.

Test Specialist

sdlc-test

Deep-dive testing for complex features.

Security Specialist

sdlc-security

Security review and vulnerability assessment.

Performance Specialist

sdlc-perf

Performance profiling and optimization.

Database Specialist

sdlc-database

Schema design and query optimization.

Root Cause Analyst

sdlc-root-cause

Diagnoses failures and broken behavior. Dual-mode: on-demand diagnosis and Phase 4 failure analysis.

Code Linter

code-linter

Hybrid linting — runs ESLint/ShellCheck for deterministic checks, then Claude analysis for logic errors and security issues.

Codebase Auditor

codebase-auditor

Audits against OWASP Top 10, CWE, SANS/CIS benchmarks, and engineering hygiene frameworks.

Firecrawl Specialist

firecrawl

Web scraping, crawling, and structured data extraction via the Firecrawl API.

Hotfix Agent

sdlc-hotfix

Structured bug fix workflow outside the feature SDLC for production issues.

Deprecation & Migration

deprecation-and-migration

Safe removal of code, dependencies, and features. Prevents breaking changes from careless removal with Chesterton's Fence principle.

Browser Testing

browser-testing-with-devtools

Live browser inspection via Chrome DevTools MCP — DOM, console, network, performance traces. Treats browser content as untrusted; constrains JavaScript execution.

Source-Driven Development

source-driven-development

Grounds framework-specific code in official documentation with citations. Detect stack → fetch docs → implement → cite. Prevents stale-training-data drift.

Documentation & ADRs

documentation-and-adrs

Architecture Decision Records and inline documentation patterns. ADR template + lifecycle (Proposed → Accepted → Superseded). Comment the why, not the what.

Stack Watch

stack-watch

Proactive vendor change monitoring. Discovers your stack from the repo, maintains a diffable STACK.md, then probes vendor changelogs for deprecations, breaking changes, and security advisories. Maps every finding back to the files it touches.

Pipeline Agents

One agent per phase, working in sequence through the pipeline.

0

Discovery Agent

sdlc-discover

Interviews you about the problem, challenges assumptions, and produces a scoped Problem Brief with clear success criteria.

  • Scope definition and constraint setting
  • Success criteria extraction
  • Risk identification
  • Feature sizing (S / M / L)
1

Architecture Agent

sdlc-architect

Reads your codebase, analyzes existing patterns, and produces a Technical Design Document that fits what's already there.

  • Codebase analysis and pattern detection
  • Schema and API design
  • Component specification
  • Technical Design Document generation
1.5

Design Agent

sdlc-design

Creates UI/UX specifications for UI-heavy features. Component specs, layout mockups, and interaction patterns.

  • Component specification
  • Layout and spacing design
  • Interaction pattern definition
  • Responsive design considerations
2

Planning Agent

sdlc-plan

Decomposes the Technical Design Document into ordered, verifiable code prompts with failure recovery strategies.

  • Task decomposition
  • Dependency ordering
  • Verification step definition
  • Failure recovery planning
3

Build Agent

sdlc-build

Executes code prompts sequentially — writes code, runs checks, handles errors, and commits with conventional messages.

  • Code generation and editing
  • Automated check execution
  • Error recovery and retry
  • Conventional commit creation
3+

Compliance Agent

sdlc-compliance

Validates code against 23 rules during build and verification. Catches issues before they reach production.

  • Code standards enforcement
  • Security pattern validation
  • API route validation
  • Type safety checks
4

Verification Agent

sdlc-verify

Runs the full QA suite — type checks, lint, tests, security review — and produces a comprehensive QA report.

  • Test execution and reporting
  • Security audit
  • Compliance audit
  • QA report generation
5

Ship Agent

sdlc-ship

Handles post-merge deployment verification, updates CLAUDE.md architecture docs, and creates release notes.

  • Deployment verification
  • CLAUDE.md update
  • Release note generation
  • Production health check
6

Retro Agent

sdlc-retro

Reviews the build cycle, identifies patterns, proposes improvements to skills and process, and tracks tech debt.

  • Build cycle analysis
  • Pattern identification
  • Skill improvement proposals
  • Tech debt tracking

Review Subagents v8

Read-only specialist for a second opinion on changed code. Invoked by other agents during verification, or directly by the user. (typescript-reviewer was retired in v9.0 in favor of the built-in /code-review — see Retired Agents below.)

Silent Failure Hunter

silent-failure-hunter

Hunts swallowed errors, floating promises, empty catches, JSON.parse without try/catch, non-Error throws, and similar patterns. Produces numbered findings with severity, impact, and remediation hints. Read-only.

Retired Agents v9.0

Claude Code caught up to these. Rather than ship agents that compete with native capabilities — and crowd the trigger registry that decides which skills fire — v9.0 retired them and points at what replaced each one.

Tech Consultant

tech-consultantRetired

Passive pattern enforcer. Its always-on rules now live in CLAUDE.md, where they genuinely load every session.

Replaced by CLAUDE.md “Pattern Enforcement” section

Context Engineer

context-engineeringRetired

Context hierarchy and packing strategies. Claude Code now manages context natively; the playbook survives as a reference file.

Replaced by native Claude Code context management

Code Simplification

code-simplificationRetired

Active refactoring workflow. Claude Code ships /simplify natively; the project-specific patterns survive as a reference file.

Replaced by the built-in /simplify command

TypeScript Reviewer

typescript-reviewerRetired

Second-opinion TS/Next.js reviewer. Claude Code's /code-review now covers this with effort levels and cloud review.

Replaced by the built-in /code-review command

Slash Commands

Operations triggered by typing /name in Claude Code. Grouped by purpose.

SDLC Flow

/new-feature

Start a new feature in Phase 0 (Discovery).

/status

Print current SDLC state — active feature, phase, branch, in-flight items.

/gate-check

Verify the current phase's gate checklist before advancing.

/resume-feature

Resume work on a previously started feature.

/skip-phase

Mark a phase complete without running it. Use sparingly.

/rollback

Revert a feature to a prior phase.

Authoring

/skill-createv8

Standardized workflow for authoring new skills — frontmatter, profile registration, CLAUDE.md update.

Session

/checkpointv8

Snapshot the current session to .claude/snapshots/checkpoint-<ts>-<label>/ with state, git context, and resume notes.

/sdlc-helpv9

See what the toolkit installed here can do — profile, skills, commands, and the phrases that trigger them.

/toolkit-doctorv9

Read-only install self-diagnosis: frontmatter validity, hook wiring, state health, manifest drift, version currency.

/whats-newv9

Upgrade brief — what changed between your installed toolkit version and the newest release.

Git

/prv8

Ad-hoc PR creation outside the Phase 5 flow. Title from commits, body from SDLC artifacts.

Lifecycle Hooks v8

Non-blocking Node scripts that fire on Claude Code events. Warnings go to stdout where the agent sees them; never blocks your work. Ships with the full install profile.

SessionStartsession-start

Prints active feature, phase, branch, and in-flight features so each session resumes with full SDLC context.

PostToolUsepost-edit-format

Silently runs Prettier on edited TS/JS/MD/CSS/JSON files when prettier is available.

PostToolUsepost-edit-console-warn

Warns when console.log/debug/info appears in production source. Skips test/scripts/config files.

PreToolUsepre-bash-commit-guard

Before any git commit, runs npm run lint + tsc --noEmit and emits a warning if either fails. Never blocks the commit.

PreCompactpre-compact

Snapshots .claude/sdlc-state.json to .claude/snapshots/sdlc-state-<ts>.json before context summarization.

Stopstop-pattern-extract

Rate-limited (5min) extractor that diffs commits since the last run and appends fix/revert pattern candidates to .claude/snapshots/pattern-candidates.md for Phase 6 retros.