Claude Code¶

Claude Code gets full learnship capabilities including real parallel subagents, specialist agent dispatch, and the complete workflow suite.

Install¶

npx learnship --claude --global

Installs to ~/.claude/learnship/.

Alternatively, install via the community marketplace (no terminal required):

/plugin marketplace add FavioVazquez/learnship-marketplace
/plugin install learnship@learnship-marketplace

Invoke commands¶

All learnship workflows use the /learnship: prefix:

/learnship:ls
/learnship:new-project
/learnship:discuss-phase 1
/learnship:plan-phase 1
/learnship:execute-phase 1
/learnship:verify-work 1
/learnship:quick "fix the login bug"
/learnship:help
/learnship:review              # v2.0: multi-persona code review
/learnship:ship                # v2.0: test → commit → push → PR
/learnship:compound            # v2.0: capture solved problem as knowledge
/learnship:challenge           # v2.0: stress-test scope
/learnship:ideate              # v2.0: codebase-grounded idea generation

Skills¶

Skills are installed as native Claude Code skills — they appear as first-class slash commands immediately after install:

~/.claude/skills/
├── agentic-learning/
│   ├── SKILL.md           ← native skill, all actions inline
│   └── references/        ← supplementary detail files
└── impeccable/
    └── SKILL.md           ← all 21 sub-skill bodies inlined

Invoke with slash commands:

/agentic-learning learn React hooks
/agentic-learning quiz
/agentic-learning either-or
/agentic-learning brainstorm my auth design
/impeccable audit
/impeccable polish
/impeccable critique

Or just work normally: skills activate at workflow checkpoints when learning_mode: "auto".

Parallel subagents¶

Claude Code supports real parallel subagents. Enable in your project:

.planning/config.json

{ "parallelization": true }

When enabled: - new-project spawns a project-researcher for domain research, then a roadmapper for phase planning - plan-phase spawns three dedicated subagents (phase-researcher, planner, plan-checker) each with a fresh 200k context budget - execute-phase dispatches each independent plan to its own executor agent: plans in the same wave run in parallel - debug spawns a dedicated debugger subagent for deep root-cause investigation - review spawns a code-reviewer subagent through 6 review lenses - challenge spawns a challenger subagent for scope stress-testing - compound spawns a solution-writer subagent to capture knowledge - ideate spawns an ideation-agent for codebase-grounded idea generation - secure-phase spawns a security-auditor for STRIDE verification - verify-work spawns debugger, planner, and verifier subagents for UAT diagnosis - validate-phase spawns a verifier subagent for retroactive test coverage - docs-update spawns a doc-writer and doc-verifier for documentation generation

All 17 agent personas have inline <persona_context> blocks as sequential fallback when parallelization is disabled.

Session hooks (v2.2)¶

Claude Code gets 4 session hooks installed automatically:

Hook	What it does
Statusline	Shows model, task/phase, directory, and a context usage bar (green → yellow → orange → red)
Context monitor	Warns the AI at 35% remaining (WARNING) and 25% remaining (CRITICAL) context to prevent unfinished work
Prompt guard	Scans `.planning/` file writes for prompt injection patterns (advisory, does not block)
Session state	Injects STATE.md orientation at session start and triggers background update checks

Hooks are installed to ~/.claude/settings.json automatically. No configuration needed.

Interactive questions (v2.2)¶

14 workflows now present user decisions via AskUserQuestion — Claude Code's native structured question tool. You'll see clickable option cards instead of plain text lists during /new-project, /settings, /discuss-phase, and other interactive workflows.

Capabilities¶

Feature	Status
Slash commands	✅ `/learnship:*` prefix
`/agentic-learning` skill	✅ Native skill (`~/.claude/skills/`)
`/impeccable` skill suite	✅ Native skill, all 21 actions inlined
Parallel subagents	✅ opt-in
Wave execution	✅ opt-in
Agent personas (17)	✅ `Task()` subagents + inline `<persona_context>`
Session hooks	✅ 4 hooks
Interactive questions	✅ `AskUserQuestion`

Tips¶

AGENTS.md is auto-loaded by Claude Code as a project rule if placed at the project root.
Slash command prefix is /learnship:: not /learnship- (that's OpenCode).
Subagents are opt-in. Default is sequential which is always safe. Enable parallelization only after your first project works end-to-end.