claudify-agent 1.0.5 → 1.0.6

package/.claude/commands/learn.md

@@ -0,0 +1,16 @@
+---
+description: Turn a mistake I just corrected into a rule so it never repeats
+argument-hint: <what went wrong and what is right>
+---
+
+Lesson to learn: $ARGUMENTS
+
+If the argument is empty, infer the lesson from the correction I made earlier in this conversation.
+
+1. State the lesson as one imperative line: what to do (or never do) and when. Add a one-line why if not obvious.
+2. Pick where it belongs, most specific first:
+   - business invariant of this project → Invariants in `spec/config.md`;
+   - a technology or practice covered by a skill → that `.claude/skills/<name>/SKILL.md`, under a `## Lessons` section;
+   - general working style → `.claude/rules/lessons.md` (always loaded, keep it under 30 lines).
+3. Check the target for an existing rule that covers it; if so, sharpen that line instead of adding a duplicate.
+4. Show me the exact diff and wait for my approval before writing.

package/.claude/commands/spec-apply.md

@@ -0,0 +1,14 @@
+---
+description: Implement the next unchecked tasks of a spec change
+argument-hint: <change-name> [task number]
+---
+
+Implement spec change: $ARGUMENTS
+
+1. Read `spec/changes/<change-name>.md` and the Invariants in `spec/config.md`. Nothing else up front; read code only as each task needs it.
+2. Take the requested task, or the first unchecked `- [ ]` task. Work one task at a time.
+3. Implement it, then run its DoD. Fix until it passes (max 5 attempts, then stop and report).
+4. Tick it `- [x]` in the change file. If reality differs from the spec, update the spec (Requirements/Design) in the same step and tell me.
+5. Report in 2-3 lines: task done, proof run, next task. Continue to the next task only if I asked for more than one.
+
+When every task is ticked, set `Status: done` and suggest `/spec-archive <change-name>`.

package/.claude/commands/spec-archive.md

@@ -0,0 +1,14 @@
+---
+description: Merge a finished change's requirements into the spec backbone and archive it
+argument-hint: <change-name>
+---
+
+Archive spec change: $ARGUMENTS
+
+1. Read `spec/changes/<change-name>.md`. Refuse if any task is unchecked; list them.
+2. For each `### <capability>` block under Requirements, update `spec/specs/<capability>.md` (create if new):
+   - Keep only Requirements and Scenarios: the current truth of the system, no history, no tasks.
+   - ADDED → append. MODIFIED → replace the requirement of the same name in full. REMOVED → delete it.
+   - If a MODIFIED/REMOVED name does not exist in the backbone, stop and ask instead of guessing.
+3. Move the change file to `spec/changes/archive/YYYY-MM-DD-<change-name>.md`.
+4. Show me the diff of `spec/specs/` in a short summary.

package/.claude/commands/spec-new.md

@@ -0,0 +1,39 @@
+---
+description: Draft a lightweight spec for a new feature (single file, ~800 words)
+argument-hint: <change-name> <what the feature should do>
+---
+
+Create a spec change for: $ARGUMENTS
+
+1. Read `spec/config.md` (stop and offer to create it from the template if missing). Read only the `spec/specs/*.md` capabilities this feature touches, and only the source-of-truth sections it needs.
+2. Ask me (one batched question) anything that changes scope. Do not guess scope.
+3. Write `spec/changes/<change-name>.md` with exactly these sections, following the rules in `spec/config.md`:
+
+```markdown
+# <change-name>
+Status: draft | Depends on: <change or none>
+
+## Why
+## Scope
+- In: ...
+- Out (non-goals): ...
+## Requirements
+<!-- One block per affected capability, named like spec/specs/<capability>.md -->
+### <capability> (new | modified)
+#### ADDED Requirement: <name>
+<statement with SHALL/MUST>
+- Scenario <name>: GIVEN ... WHEN ... THEN ...
+#### MODIFIED Requirement: <existing name>
+<full new text, not just the diff>
+#### REMOVED Requirement: <existing name>
+<reason>
+## Design
+<only decisions not obvious from the code: data flow, schema, API, edge cases. Omit section if trivial.>
+## Risks
+## Tasks
+- [ ] 1. <task> — files: ... — DoD: ...
+```
+
+4. Keep the whole file under ~800 words, plus ~300 per extra capability. No restating config, docs or code; link them.
+   If it still does not fit (e.g. more than 4 capabilities), propose splitting into smaller changes with `Depends on:` instead of one giant spec.
+5. Do not write code. Finish by showing the task list and asking me to approve before `/spec-apply`.

package/CLAUDE.md

@@ -20,6 +20,11 @@
   - Validation steps: `.claude/rules/plan-mode-validation.md`
   - Plan structure guide: `.claude/rules/plan-mode-structure.md`
 
+## 2b. 📐 Spec Workflow (features only, opt-in by command)
+- `/spec-new` → `/spec-apply` → `/spec-archive`. Specs live in `spec/` (config, changes, specs backbone).
+- Bugs and small tasks: no spec, work directly.
+- After the user corrects a mistake, suggest `/learn` so it never repeats.
+
 ## 3. 🧪 Zero-Token Local Verification Loop
 After modifying code, execute local verification based on project stack:
 - **Node/TS/NestJS:** `npm run lint:fix && npm run typecheck`

package/bin/cli.js

@@ -41,6 +41,15 @@ if (command === 'init') {
       console.log(`  ⚠️  Kept your existing .claude/settings.json — merge the "hooks" block from ${settingsSrc} manually.`);
     }
 
+    // 3. Scaffold spec/ for the feature workflow (never overwrite an existing config)
+    const specConfig = join(targetDir, 'spec', 'config.md');
+    if (existsSync(specConfig)) {
+      console.log('  ⚠️  Kept your existing spec/config.md');
+    } else {
+      await cp(join(packageRoot, 'templates', 'spec', 'config.md'), specConfig);
+      console.log('  ✅ Created spec/config.md (fill in context & invariants, then use /spec-new)');
+    }
+
     console.log('\n🎉 Successfully initialized Claudify!');
     console.log('👉 Next Steps:');
     console.log('   - Review CLAUDE.md in your project root.');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudify-agent",
-  "version": "1.0.5",
+  "version": "1.0.6",
   "description": "Enterprise-grade workflow & context optimization framework for Claude Code, Claude Desktop, and AI coding agents.",
   "repository": {
     "type": "git",
@@ -16,6 +16,7 @@
     "CLAUDE.md",
     ".claude",
     "bin",
+    "templates",
     "README.md",
     "package.json"
   ],

package/templates/spec/config.md

@@ -0,0 +1,27 @@
+# Spec Config
+
+> Read by `/spec-new` and `/spec-apply` only. Keep it short: every line here is paid for on each spec command.
+
+## Context
+- Project: <one or two lines: what the system does>
+- Stack: <languages, frameworks>
+- Sources of truth (read before writing a change): <design docs, ADRs>. If a change contradicts them, stop and say so.
+
+## Invariants
+<!-- Rules every change MUST respect. One line each. -->
+- <e.g. money amounts are integers in base units, never floats>
+
+## Style
+- Language: <e.g. Vietnamese prose, English technical terms and spec keywords>
+- Concise: bullets, tables, short snippets. Reference docs by section instead of copying them.
+
+## Rules
+### Proposal
+- Must have Why, Scope (in / out = Non-goals), Depends on, Affected capabilities, Risks.
+### Requirements
+- `### Requirement: <name>` containing SHALL or MUST.
+- Each has `#### Scenario:` GIVEN / WHEN / THEN with concrete values: at least one happy path and one failure.
+### Tasks
+- Each task fits in 2-4 hours, lists files/packages touched and its DoD.
+- Tests ship with the task that adds the logic, not at the end.
+- DoD default: <e.g. `npm run lint && npm run typecheck && npm test`>