@slamb2k/mad-skills 2.0.24 → 2.0.25

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "mad-skills",
   "description": "AI-assisted planning, development and governance tools",
-  "version": "2.0.24",
+  "version": "2.0.25",
   "author": {
     "name": "slamb2k",
     "url": "https://github.com/slamb2k"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@slamb2k/mad-skills",
-  "version": "2.0.24",
+  "version": "2.0.25",
   "description": "Claude Code skills collection — full lifecycle development tools",
   "type": "module",
   "repository": {

package/skills/build/SKILL.md

@@ -1,8 +1,8 @@
 ---
 name: build
 description: Context-isolated feature development pipeline. Takes a detailed design/plan as argument and executes the full feature-dev lifecycle (explore, question, architect, implement, review, ship) inside subagents so the primary conversation stays compact. Use when you have a well-defined plan and want autonomous execution with minimal context window consumption.
-argument-hint: <detailed design/plan to implement> [--skip-questions] [--skip-review] [--no-ship] [--parallel-impl]
-allowed-tools: Bash, Read, Write, Edit, Glob, Grep, Agent, AskUserQuestion
+argument-hint: <plan text or spec file path> [--skip-questions] [--skip-review] [--no-ship] [--parallel-impl]
+allowed-tools: Bash, Read, Write, Edit, Glob, Grep, Agent, AskUserQuestion, TaskCreate, TaskUpdate, TaskGet, TaskList
 ---
 
 # Build - Context-Isolated Feature Development
@@ -103,16 +103,14 @@ For each row, in order:
 
 1. Capture **PLAN** (the user's argument) and **FLAGS**
 2. **Load project context** — invoke `/prime` to load domain-specific context
-   (CLAUDE.md, goals, specs, memory). If /prime is unavailable, fall back to
-   manually scanning CLAUDE.md and goals/ directory.
+   (CLAUDE.md, specs, memory). If /prime is unavailable, fall back to
+   manually scanning CLAUDE.md and specs/ directory.
 3. Detect project type using `references/project-detection.md` to populate
    **PROJECT_CONFIG** (language, test_runner, test_setup)
-3. Check for outstanding questions from previous work:
+3. Check for outstanding items from previous work:
+   - Query persistent tasks via `TaskList` for incomplete items
    - Search CLAUDE.md for a "Known Issues" or "Open Questions" section
-   - Search `goals/` for files containing "open_question" or "unresolved"
-   - Search memory (if available) for recent items of type "task" or
-     "open_question" that are unresolved
-   - Check for `DEBRIEF_ITEMS` in any recent build logs
+   - Search memory (if available) for recent unresolved items
 4. If outstanding items found, present via AskUserQuestion:
    ```
    "Found {count} outstanding items from previous work:"
@@ -125,7 +123,23 @@ For each row, in order:
      Items marked "incorporate" get appended to the PLAN as additional
      requirements for Stage 1 to explore.
    - **"No, proceed with the build"** → continue normally
-5. Create a task list tracking all stages
+---
+
+## Plan Resolution
+
+Before Stage 1, resolve the PLAN argument into content:
+
+1. **File detection** — If the argument contains `/` or ends with
+   `.md`, `.yaml`, `.json`, or `.txt`, treat it as a file path:
+   - Try reading the path as-is
+   - If not found, try `specs/{arg}`
+   - If found, use file content as PLAN
+   - If not found at any location, treat the original argument as free-form text
+2. **Free-form text** — If not a file path (or file not found), use the argument
+   verbatim as PLAN
+3. **Display** — In the Input box, show the resolved source:
+   - File: `Plan: {file path} ({line count} lines)`
+   - Text: `Plan: inline ({word count} words)`
 
 ---
 
@@ -303,8 +317,9 @@ Invoke the `/ship` skill:
    Options:
    - **"Fix now"** → create a task list of resolution activities for
      each item; present for user confirmation, then work through them
-   - **"Create tasks for future sessions"** → create persistent tasks via
-     `TaskCreate` for each item, or append to CLAUDE.md as Known Issues
+   - **"Create tasks for future sessions"** → use `TaskCreate` for each
+     item as a persistent task, with category as prefix and suggested
+     action as description
    - **"Note and continue"** → acknowledge items without formal tracking;
      log to memory (if exists) or as source file comments. No further action.
    - **"Let me choose per item"** → present each individually with full
@@ -337,7 +352,7 @@ Invoke the `/ship` skill:
 │     Findings addressed: {count fixed} / {count found}
 │
 │  📊 Debrief: {count resolved} / {count surfaced}
-│     {list of created goals/tasks}
+│     {list of created tasks}
 │
 │  🔗 Links
 │     PR:  {pr_url}

package/skills/build/references/stage-prompts.md

@@ -385,9 +385,10 @@ Options:
      for user confirmation, then work through them.
 
 - "Create tasks for future sessions"
-  → For each item, create a persistent task via TaskCreate with the
-     suggested_action as the task description, or append to CLAUDE.md
-     as a "Known Issues" section. Group by category, not one per item.
+  → For each item, use TaskCreate to create a persistent task with:
+     title: "[{category}] {summary}"
+     description: "{description}\n\nEvidence: {evidence}\nSuggested action: {suggested_action}"
+     Format: one task per item for granular tracking.
 
 - "Note and continue"
   → Acknowledge items without formal tracking. Log to memory (if memory
@@ -398,8 +399,9 @@ Options:
   → Present each item individually via AskUserQuestion. Show the full
      `description`, `evidence`, and `impact` fields so the user has
      enough context to decide. Options per item:
-     "Fix now" / "Add to goals" / "Explain more" / "Note and continue"
+     "Fix now" / "Create task" / "Explain more" / "Note and continue"
      - "Fix now": launch a general-purpose subagent to resolve
+     - "Create task": use TaskCreate to persist for future sessions
      - "Explain more": read the relevant source files/reports cited in
        `evidence`, summarise the context, then re-present the same item
        with the expanded explanation

package/skills/build/tests/evals.json

@@ -17,6 +17,14 @@
       { "type": "semantic", "value": "Describes a multi-stage development lifecycle that includes exploration, architecture design, implementation, code review, and shipping, with stages running in isolated subagents" }
     ]
   },
+  {
+    "name": "file-path-detection",
+    "prompt": "Build specs/user-auth.md",
+    "assertions": [
+      { "type": "contains", "value": "██" },
+      { "type": "semantic", "value": "Detects that the argument is a file path (contains / or ends with .md) and attempts to read it as a spec file rather than treating it as literal plan text. Shows the resolved file path in the Input display." }
+    ]
+  },
   {
     "name": "flag-handling",
     "prompt": "Build --skip-questions --no-ship: Refactor the auth module to use middleware pattern",

package/skills/manifest.json

@@ -1,5 +1,5 @@
 {
-  "generated": "2026-03-11T04:28:37.105Z",
+  "generated": "2026-03-11T04:39:55.849Z",
   "count": 10,
   "skills": [
     {
@@ -16,7 +16,7 @@
       "name": "build",
       "directory": "build",
       "description": "Context-isolated feature development pipeline. Takes a detailed design/plan as argument and executes the full feature-dev lifecycle (explore, question, architect, implement, review, ship) inside subagents so the primary conversation stays compact. Use when you have a well-defined plan and want autonomous execution with minimal context window consumption.",
-      "lines": 363,
+      "lines": 378,
       "hasScripts": false,
       "hasReferences": true,
       "hasAssets": false,
@@ -55,7 +55,7 @@
     {
       "name": "prime",
       "directory": "prime",
-      "description": "\"Load project context before implementing features or making architectural decisions. Invoke proactively at the start of significant work on any project. Scans CLAUDE.md, README, goals/, specs/, docs/, and source structure to build a context summary. Supports optional domain hints to focus on specific areas of the codebase. Use when you need project conventions, architecture understanding, or domain context before coding.\"",
+      "description": "\"Load project context before implementing features or making architectural decisions. Invoke proactively at the start of significant work on any project. Scans CLAUDE.md, README, specs/, docs/, and source structure to build a context summary. Supports optional domain hints to focus on specific areas of the codebase. Use when you need project conventions, architecture understanding, or domain context before coding.\"",
       "lines": 148,
       "hasScripts": false,
       "hasReferences": true,
@@ -86,7 +86,7 @@
       "name": "speccy",
       "directory": "speccy",
       "description": "Deep-dive interview skill for creating comprehensive specifications. Reviews existing code and docs, then interviews the user through multiple rounds of targeted questions covering technical implementation, UI/UX, concerns, and tradeoffs. Produces a structured spec via create-specification. Use when starting a new feature, system, or major change that needs a spec.",
-      "lines": 232,
+      "lines": 245,
       "hasScripts": false,
       "hasReferences": true,
       "hasAssets": false,

package/skills/prime/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: prime
-description: "Load project context before implementing features or making architectural decisions. Invoke proactively at the start of significant work on any project. Scans CLAUDE.md, README, goals/, specs/, docs/, and source structure to build a context summary. Supports optional domain hints to focus on specific areas of the codebase. Use when you need project conventions, architecture understanding, or domain context before coding."
+description: "Load project context before implementing features or making architectural decisions. Invoke proactively at the start of significant work on any project. Scans CLAUDE.md, README, specs/, docs/, and source structure to build a context summary. Supports optional domain hints to focus on specific areas of the codebase. Use when you need project conventions, architecture understanding, or domain context before coding."
 argument-hint: "[domain hints: comma-separated directory or topic names to focus on]"
 allowed-tools: Read, Glob, Grep, LS, Agent
 ---
@@ -66,7 +66,7 @@ subagent — the primary thread only sees a structured PRIME_REPORT.
 
 Extract domain hints from the request (comma-separated). These are directory
 names or topic keywords to focus the context scan on. If no domain specified,
-load core context only (CLAUDE.md, README, goals/, specs/).
+load core context only (CLAUDE.md, README, specs/).
 
 ## Step 2: Load Context via Subagent
 
@@ -92,7 +92,7 @@ Limit PRIME_REPORT to 30 lines maximum.
 
 1. CLAUDE.md — Project conventions, architecture, instructions
 2. README.md — Project overview, setup, usage
-3. goals/ or specs/ — Project goals, specs, roadmap (scan directory if present)
+3. specs/ — Project specifications and roadmap (scan directory if present)
 4. docs/ — Documentation directory (scan if present)
 
 If a file is missing, record as NOT FOUND and continue.

package/skills/speccy/SKILL.md

@@ -90,12 +90,12 @@ Before asking any questions, build a thorough understanding of the project:
 
 1. **Capture GOAL** — the user's argument describing what needs to be specified
 2. **Load project context** — invoke `/prime` to load domain-specific context
-   (CLAUDE.md, goals, specs, memory). If /prime is unavailable, fall back to
+   (CLAUDE.md, specs, memory). If /prime is unavailable, fall back to
    the manual scan below.
 3. **Scan the project** (skip items already loaded by /prime):
    - Read `CLAUDE.md` if present (project conventions, structure, domain)
-   - Read spec directory: `specs/` for existing specifications
-   - Scan existing specs and design docs for context
+   - Scan `specs/` directory for existing specifications
+   - Scan existing design docs for context
    - Read relevant source code that relates to the GOAL
    - Check memory for prior decisions or open questions related to the GOAL
 3. **Identify knowledge gaps** — what must you learn from the user to write
@@ -163,19 +163,28 @@ proceeding to spec generation.
 
 Once the interview is complete and decisions are confirmed:
 
-1. **Invoke the `create-specification` skill** using the Skill tool:
+1. **Create `specs/` directory** if it doesn't exist:
+   ```
+   mkdir -p specs
+   ```
+
+2. **Invoke the `create-specification` skill** using the Skill tool:
    ```
    Skill(skill: "create-specification")
    ```
 
-2. **Provide the full context** to the skill as input:
+3. **Provide the full context** to the skill as input:
    - The original GOAL
    - All decisions from the interview rounds
    - Any relevant code/architecture context discovered in Stage 1
    - The spec purpose should match the GOAL description
+   - **Instruct the skill to save the spec to `specs/{name}.md`** where
+     `{name}` is a kebab-case slug derived from the GOAL (e.g.,
+     `specs/user-auth.md`, `specs/payment-integration.md`)
 
-3. The `create-specification` skill will handle formatting, naming, and
-   saving the spec file according to its own template and conventions.
+4. The `create-specification` skill will handle formatting and template
+   structure. The `specs/` directory is the standard location — `/build`
+   and `/prime` both scan it automatically.
 
 ---
 
@@ -201,6 +210,10 @@ After the spec is created, report to the user:
 │  🔗 Links
 │     Spec: {spec file path}
 │
+│  ⚡ Next steps
+│     1. Review the spec: {path}
+│     2. Run `/build {spec path}` to implement (reads the file automatically)
+│
 └─────────────────────────────────────────────────
 ```