The 7 Laws of AI Agent Discipline

Stop your Claude Code agent from skipping steps, guessing, and declaring "done" without verifying.

New here? → QUICKSTART.md (2 minutes)

---

Install

Option A — Claude Code plugin (recommended, 2 commands)

Run inside the Claude Code CLI:

/plugin marketplace add naimkatiman/continuous-improvement
/plugin install continuous-improvement@continuous-improvement

Update later with /plugin marketplace update continuous-improvement. Browse the other 8 PM plugins with /plugin (interactive picker) or see the full list below.

Option B — npm CLI installer

Requires Node 18/20/22 and bash (Git Bash or WSL on Windows).

# Beginner — 7 Laws skill, hooks, slash commands
npx continuous-improvement install

# Expert — adds MCP server (12 tools) + session hooks
npx continuous-improvement install --mode expert

# Optional starter packs
npx continuous-improvement install --pack react|python|go

Verify either path with /discipline or /dashboard in Claude Code.

---

The 7 Laws

#	Law	Without it, agents...
1	Research Before Executing	reinvent what already exists
2	Plan Is Sacred	scope-creep and overbuild
3	One Thing at a Time	stack untested changes
4	Verify Before Reporting	lie about being "done"
5	Reflect After Sessions	repeat the same failures
6	Iterate One Change	debug 5 changes at once
7	Learn From Every Session	lose knowledge when context ends

Research -> Plan -> Execute (one thing) -> Verify -> Reflect -> Learn -> Iterate

Full spec, reflection-block format, and anti-examples: SKILL.md.

---

Mulahazah: Auto-Leveling Learning

Hooks capture every tool call. After ~20 observations, Claude analyzes patterns and creates instincts with confidence scores:

• < 0.5 silent (stored, not surfaced)
• 0.5–0.69 suggested inline when relevant
• 0.7+ auto-applied
• User corrections drop confidence by 0.1; unused instincts decay
• Project-scoped, promoted to global after seen across 2+ projects

---

Slash Commands

/continuous-improvement    Reflect, analyze, show status
/planning-with-files       Create task_plan.md, findings.md, progress.md
/discipline                Quick reference card of the 7 Laws
/dashboard                 Visual instinct health dashboard

In expert mode, the same planning workflow is also available programmatically through the MCP tools ci_plan_init (initialize task_plan.md, findings.md, progress.md in the project root) and ci_plan_status (summarize their current contents).

---

Bundled Skills & Plugins

The default install only deploys the core 7 Laws skill. Everything below is opt-in — pick what you want.

Featured companion — the recommended pairing for the 7 Laws

⭐ proceed-with-the-recommendation is the execution arm of this repo. When you say "do all of it" / "proceed with your recommendation" / "yes do it", this skill walks the list top-to-bottom under the 7 Laws, routes each item to the right specialist (superpowers:*, ralph, workspace-surface-audit, simplify, security-review, schedule, loop), falls back to inline behavior when a specialist isn't installed, verifies per item, and halts on needs-approval instead of barreling through. If you adopt one companion alongside the core skill, adopt this one — it's purpose-built to operationalize the laws end-to-end.

Install it on its own:

SKILL=proceed-with-the-recommendation
mkdir -p ~/.claude/skills/$SKILL
curl -L https://raw.githubusercontent.com/naimkatiman/continuous-improvement/main/skills/$SKILL.md \
  -o ~/.claude/skills/$SKILL/SKILL.md

…or get it (and the rest below) by installing the continuous-improvement plugin from the marketplace.

Tier 1 — recommended pairing for beginner install

These add concrete enforcement to the 7 Laws and ship in every plugin install.

Skill	What it does	Pairs with
para-memory-files	File-based persistent memory using PARA (Projects/Areas/Resources/Archives) for cross-session context	Law 5 + Law 7
verification-loop	Six-phase verification (build, types, lint, tests, security, diff) with PASS/FAIL report	Law 4
gateguard	PreToolUse fact-forcing gate that blocks Edit/Write/destructive Bash until concrete investigation is presented	Law 1 + Law 3
tdd-workflow	RED→GREEN→REFACTOR enforcement with 80%+ coverage gate across unit/integration/E2E	Law 3 + Law 4

Tier 2 — additional skills for expert install

Layer on top of tier-1 for autonomous-mode safety, response-depth control, and context-window discipline.

Skill	What it does
safety-guard	Three-mode runtime guard (careful/freeze/guard) that blocks destructive commands and locks edits to a directory
token-budget-advisor	Heuristic input/output token estimator that offers 25%/50%/75%/100% depth choices before answering
strategic-compact	PreToolUse hook that suggests /compact at logical phase boundaries instead of arbitrary auto-compaction
wild-risa-balance	Decision-framing lens that pairs WILD (bold) generation with RISA (safe) execution; splits recommendation lists into pilots above a baseline so bold options stop losing to safe ones in a flat list

The /learn-eval slash command also ships with the expert install: extract a session pattern, run a checklist quality gate, and decide global-vs-project save location before writing any skill file.

Always-bundled companion skills (skills/)

Drop-in single-file skills, copy to ~/.claude/skills/<name>/SKILL.md.

Skill	What it does	Source
ralph	Autonomous loop that executes a PRD story-by-story with quality checks between iterations	snarktank/ralph
superpowers	Activates task-appropriate skills automatically (brainstorming, git-worktrees, TDD, code review, etc.)	obra/superpowers
workspace-surface-audit	Audits the active repo, MCP servers, plugins, and env, then recommends high-value skills/workflows	continuous-improvement

Plugin marketplace (plugins/)

Install via /plugin marketplace add naimkatiman/continuous-improvement then /plugin install <name>@continuous-improvement (the -dev suffix is only used for the local in-repo development marketplace at plugins/continuous-improvement/.claude-plugin/marketplace.json).

Plugin	Skills	Focus
continuous-improvement	4	The 7 Laws + Mulahazah (this repo's core plugin bundle)
pm-product-discovery	12	Ideation, experiments, assumption testing, feature prioritization, interview synthesis
pm-product-strategy	12	Vision, lean canvas, business model, SWOT, PESTLE, Ansoff, Porter's Five Forces, monetization
pm-execution	15	PRDs, OKRs, roadmaps, sprints, pre-mortems, user stories, retros, release notes
pm-market-research	7	Personas, market segments, market sizing, journey maps, sentiment, competitor analysis
pm-data-analytics	3	SQL query generation, cohort analysis, A/B test analysis
pm-go-to-market	6	GTM strategy, growth loops, GTM motions, beachhead segments, ICP, battlecards
pm-marketing-growth	5	Marketing ideas, value props, North Star metrics, product naming, positioning
pm-toolkit	4	Resume review, NDA drafting, privacy policy, grammar check

PM plugins by Paweł Huryn.

---

GitHub Action: Transcript Linter

Lint agent behavior in CI. Detects skipped laws.

- uses: naimkatiman/continuous-improvement@v3
  with:
    transcript-path: agent-log.jsonl
    strict: true

Catches: writes without prior research (Law 1), too many edits without verification (Law 3), code changes without tests/builds (Law 4), too many files at once (Law 6). Run locally with node bin/lint-transcript.mjs <file>.

---

Uninstall

npx continuous-improvement install --uninstall

Removes skill, hooks, commands, MCP server. Learned instincts in ~/.claude/instincts/ are preserved — delete manually for a clean slate.

---

More

• QUICKSTART.md — 2-minute setup
• SKILL.md — full 7 Laws spec
• examples/ — bug fix, feature build, refactor walkthroughs
• CONTRIBUTING.md — architecture, repo internals
• SECURITY.md

MIT.