claude-code-modes — Roadmap

claude-code-modes — Roadmap

Solo AI-assisted build. Phases are chunky — each completable in one session. Full test suite throughout.

---

Phase 1: Project Scaffold + Base Prompt Fragments

**Goal:** Project compiles and assembles a complete base prompt (the `none` mode) from markdown fragments with dynamic env info.

**Build:**

• package.json with bun, typescript config, test setup
• tsconfig.json
• src/types.ts — Agency, Quality, Scope enums, preset type, CLI args type
• src/env.ts — shell exec calls for CWD, git status, platform, shell, OS version, git log
• src/assemble.ts — reads markdown fragments from prompts/, substitutes {{VAR}} template variables, concatenates in order, writes temp file
• prompts/base/intro.md — identity + cyber risk (extracted from Claude Code getSimpleIntroSection)
• prompts/base/system.md — tool permissions, hooks, tags, compression (extracted from getSimpleSystemSection)
• prompts/base/doing-tasks.md — universal KEEP-only instructions from audit (extracted from getSimpleDoingTasksSection)
• prompts/base/actions-autonomous.md — softened risky-action guidance
• prompts/base/actions-cautious.md — full risky-action guidance (near-verbatim from getActionsSection)
• prompts/base/tools.md — tool usage preferences (extracted from getUsingYourToolsSection)
• prompts/base/tone.md — style guidelines, stripped of behavioral items (extracted from getSimpleToneAndStyleSection)
• prompts/base/session-guidance.md — AskUserQuestion, ! command, agent tool, skills (extracted from getSessionSpecificGuidanceSection)
• prompts/base/env.md — template with all {{VAR}} placeholders (mirrors computeSimpleEnvInfo)
• prompts/modifiers/context-pacing.md — the "take your time, pause gracefully" section
• Unit tests: env detection returns expected shapes, assemble reads fragments and substitutes variables correctly, template variables are all replaced (no {{ remains in output)

**Test checkpoint:** `bun test` — all unit tests pass. `bun run src/assemble.ts` with a mock config produces a complete prompt file with no unresolved template variables. Diff the output against Claude Code's actual system prompt to verify base sections match.

---

Phase 2: Axis Fragments + Presets

**Goal:** All 9 axis fragments and 5 presets exist. Assembly can produce a mode-specific prompt by combining base + axis selections.

**Build:**

• prompts/axis/agency/autonomous.md — high initiative, free local actions, architectural decisions without asking
• prompts/axis/agency/collaborative.md — explains reasoning, checks in, confirms before large changes
• prompts/axis/agency/surgical.md — minimal blast radius, precise execution
• prompts/axis/quality/architect.md — proper abstractions, error handling, forward-thinking, thorough output
• prompts/axis/quality/pragmatic.md — match existing patterns, incremental improvement, balanced output
• prompts/axis/quality/minimal.md — smallest correct change,