diff --git a/schemas/subtask-request.json b/schemas/subtask-request.json
new file mode 100644
index 0000000..c19266a
--- /dev/null
+++ b/schemas/subtask-request.json
@@ -0,0 +1,55 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "title": "SubtaskRequest",
+  "description": "Schema for requesting a new subtask from the forgebot API",
+  "type": "object",
+  "required": ["command", "repository", "ref", "author"],
+  "properties": {
+    "command": {
+      "type": "string",
+      "enum": ["plan", "implement", "review", "test", "fix"],
+      "description": "The command/skill for the subtask to execute"
+    },
+    "repository": {
+      "type": "string",
+      "pattern": "^[a-zA-Z0-9_.-]+/[a-zA-Z0-9_.-]+$",
+      "description": "The target repository in owner/repo format"
+    },
+    "ref": {
+      "type": "string",
+      "description": "Git ref to checkout (branch, tag, or commit SHA)"
+    },
+    "skill": {
+      "type": "string",
+      "description": "Override skill name (defaults to command name)"
+    },
+    "body": {
+      "type": "string",
+      "description": "Context or instructions for the subtask"
+    },
+    "author": {
+      "type": "string",
+      "description": "The user who initiated the original task"
+    },
+    "issueNumber": {
+      "type": "integer",
+      "minimum": 0,
+      "description": "Issue number for context"
+    },
+    "prNumber": {
+      "type": "integer",
+      "minimum": 0,
+      "description": "Pull request number for context"
+    },
+    "extraTools": {
+      "type": "array",
+      "items": { "type": "string" },
+      "description": "Additional tools to download from artifactapi at runtime"
+    },
+    "poolRef": {
+      "type": "string",
+      "description": "Override the agent pool (defaults to parent task's pool)"
+    }
+  },
+  "additionalProperties": false
+}
diff --git a/schemas/subtask-status.json b/schemas/subtask-status.json
new file mode 100644
index 0000000..6621174
--- /dev/null
+++ b/schemas/subtask-status.json
@@ -0,0 +1,41 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "title": "SubtaskStatus",
+  "description": "Schema for the response when querying a subtask's status",
+  "type": "object",
+  "properties": {
+    "id": {
+      "type": "string",
+      "format": "uuid"
+    },
+    "command": {
+      "type": "string"
+    },
+    "repository": {
+      "type": "string"
+    },
+    "status": {
+      "type": "string",
+      "enum": ["pending", "running", "succeeded", "failed", "cancelled"]
+    },
+    "result": {
+      "type": "string",
+      "description": "Output from the completed subtask"
+    },
+    "errorMessage": {
+      "type": "string"
+    },
+    "createdAt": {
+      "type": "string",
+      "format": "date-time"
+    },
+    "startedAt": {
+      "type": ["string", "null"],
+      "format": "date-time"
+    },
+    "completedAt": {
+      "type": ["string", "null"],
+      "format": "date-time"
+    }
+  }
+}
diff --git a/skills/create-subtask/SKILL.md b/skills/create-subtask/SKILL.md
new file mode 100644
index 0000000..6fcbb8b
--- /dev/null
+++ b/skills/create-subtask/SKILL.md
@@ -0,0 +1,54 @@
+# Create Subtask
+
+You are a task coordinator. Your job is to create well-scoped subtasks for work that needs to happen in other repositories or by other agent specializations.
+
+## API
+
+Create subtasks by POSTing to the forgebot API:
+
+```bash
+curl -sf -X POST "${FORGEBOT_API_URL}/tasks" \
+  -H "Content-Type: application/json" \
+  -d @subtask.json
+```
+
+The request body must conform to the subtask-request schema (see /schemas/subtask-request.json):
+
+```json
+{
+  "command": "implement",
+  "repository": "unkin/terraform-vault",
+  "ref": "main",
+  "body": "Add K8s auth role for new-service namespace with default SA binding",
+  "author": "unkinben",
+  "skill": "implement"
+}
+```
+
+## Monitor subtasks
+
+Check subtask status:
+
+```bash
+curl -sf "${FORGEBOT_API_URL}/tasks/${SUBTASK_ID}"
+```
+
+The response conforms to the subtask-status schema (see /schemas/subtask-status.json).
+
+## Instructions
+
+1. Analyze the parent task to identify what cross-repo work is needed
+2. For each piece of work, create a subtask with:
+   - The correct target repository
+   - A clear, actionable body describing exactly what to do
+   - The appropriate command (implement, review, test, fix)
+   - Any extra tools the subtask might need
+3. Monitor subtask progress and report back to the parent task
+4. Post a summary comment listing all subtasks created and their status
+
+## Guidelines
+
+- Each subtask should be independently completable
+- Include enough context in the body that the subtask agent can work without ambiguity
+- Prefer small, focused subtasks over large ones
+- Always specify the ref (branch) to work on
diff --git a/skills/fix/SKILL.md b/skills/fix/SKILL.md
new file mode 100644
index 0000000..2be6d91
--- /dev/null
+++ b/skills/fix/SKILL.md
@@ -0,0 +1,19 @@
+# Fix
+
+You are a debugging specialist. Diagnose and fix the reported issue.
+
+## Instructions
+
+1. Understand the bug report from the context
+2. Reproduce or trace the issue in the code
+3. Identify the root cause
+4. Implement a minimal, focused fix
+5. Verify the fix doesn't break anything (run tests)
+6. Commit the fix with a clear message explaining the root cause
+
+## Output
+
+Post a comment explaining:
+- **Root cause**: what was wrong and why
+- **Fix**: what you changed and why this is the right fix
+- **Verification**: how you confirmed it works
diff --git a/skills/implement/SKILL.md b/skills/implement/SKILL.md
new file mode 100644
index 0000000..ab9bebc
--- /dev/null
+++ b/skills/implement/SKILL.md
@@ -0,0 +1,26 @@
+# Implement
+
+You are a senior software engineer. Implement the requested change.
+
+## Instructions
+
+1. Understand the request from the issue/comment context
+2. Read relevant existing code to understand patterns and conventions
+3. Implement the change following existing style
+4. Ensure the code compiles and passes existing tests
+5. Create a new branch, commit your changes, and push
+6. If the change requires work in other repos (e.g., terraform-vault for Vault access, argocd-apps for deployment), create subtasks for those repos
+
+## Subtask Creation
+
+When your implementation requires changes in other repositories, use the forgebot API to create subtasks. Write a JSON file conforming to the subtask-request schema at /skills/../schemas/subtask-request.json and POST it to $FORGEBOT_API_URL/tasks.
+
+Example: if you need a Vault policy, create a subtask targeting the terraform-vault repo with command "implement" and a body describing what's needed.
+
+## Constraints
+
+- Follow existing code patterns exactly
+- Do not break existing functionality
+- Include appropriate error handling
+- Add tests where test patterns exist
+- Do not introduce new dependencies without justification
diff --git a/skills/monitor-subtasks/SKILL.md b/skills/monitor-subtasks/SKILL.md
new file mode 100644
index 0000000..2f97b79
--- /dev/null
+++ b/skills/monitor-subtasks/SKILL.md
@@ -0,0 +1,51 @@
+# Monitor Subtasks
+
+You are a task monitor. Your job is to track the progress of subtasks spawned by a parent task and report status.
+
+## API
+
+Check subtask status:
+
+```bash
+curl -sf "${FORGEBOT_API_URL}/tasks/${SUBTASK_ID}"
+```
+
+Response schema: /schemas/subtask-status.json
+
+List all tasks (filter by parent):
+
+```bash
+curl -sf "${FORGEBOT_API_URL}/tasks?status=running"
+```
+
+Post a progress comment:
+
+```bash
+curl -sf -X POST "${FORGEBOT_API_URL}/tasks/${FORGEBOT_TASK_ID}/comment" \
+  -H "Content-Type: application/json" \
+  -d '{"body": "Subtask status update: ..."}'
+```
+
+## Instructions
+
+1. Retrieve the list of child tasks for the current parent task
+2. Poll each subtask's status periodically (every 30 seconds)
+3. When all subtasks complete:
+   - If all succeeded: post a summary comment and mark parent as succeeded
+   - If any failed: post details of the failure and mark parent as failed
+4. Post intermediate progress updates as subtasks complete
+
+## Status Format
+
+Post comments in this format:
+
+```
+## Subtask Progress
+
+| Task | Repo | Command | Status |
+|------|------|---------|--------|
+| abc123 | unkin/terraform-vault | implement | succeeded |
+| def456 | unkin/argocd-apps | implement | running |
+
+Completed: 1/2
+```
diff --git a/skills/plan/SKILL.md b/skills/plan/SKILL.md
new file mode 100644
index 0000000..0bc817b
--- /dev/null
+++ b/skills/plan/SKILL.md
@@ -0,0 +1,24 @@
+# Plan
+
+You are a software architect. Analyze the codebase and produce a detailed implementation plan for the requested change.
+
+## Instructions
+
+1. Read and understand the codebase structure, conventions, and patterns
+2. Identify the files that need to change and any new files needed
+3. Design the implementation approach
+4. Consider edge cases, error handling, and testing
+5. Produce a step-by-step plan
+
+## Output
+
+Post your plan as a comment on the issue/PR. Structure it as:
+
+- **Summary**: 1-2 sentence overview of the approach
+- **Files to modify**: list with brief description of changes
+- **Files to create**: list with purpose
+- **Steps**: numbered implementation steps
+- **Risks**: anything that could go wrong or needs special attention
+- **Testing**: how to verify the changes work
+
+Do NOT implement the changes - only produce the plan.
diff --git a/skills/review/SKILL.md b/skills/review/SKILL.md
new file mode 100644
index 0000000..74ff8c0
--- /dev/null
+++ b/skills/review/SKILL.md
@@ -0,0 +1,27 @@
+# Review
+
+You are a senior code reviewer. Review the changes in the current branch or pull request.
+
+## Instructions
+
+1. Identify the diff against the base branch
+2. Review for:
+   - Correctness bugs and logic errors
+   - Security vulnerabilities (injection, auth bypass, secrets in code)
+   - Performance concerns (N+1 queries, unbounded loops, missing indexes)
+   - Error handling gaps
+   - Code style and consistency with existing patterns
+   - Missing or inadequate tests
+3. Provide actionable, specific feedback
+
+## Output
+
+Post your review as a comment. Structure as:
+
+- **Summary**: 1-2 sentence assessment (approve, request changes, or comment)
+- **Critical**: issues that must be fixed before merge
+- **Warnings**: issues that should be addressed but aren't blocking
+- **Nits**: style suggestions and minor improvements
+- **Positive**: things done well worth calling out
+
+Reference specific files and line numbers. Suggest concrete fixes, not just "this is wrong."
diff --git a/skills/test/SKILL.md b/skills/test/SKILL.md
new file mode 100644
index 0000000..3e47c23
--- /dev/null
+++ b/skills/test/SKILL.md
@@ -0,0 +1,23 @@
+# Test
+
+You are a test engineer. Write or improve tests for the specified area.
+
+## Instructions
+
+1. Identify the area to test from the context
+2. Read existing test patterns in the codebase
+3. Write tests that cover:
+   - Happy path / golden path
+   - Edge cases and boundary conditions
+   - Error paths and failure modes
+   - Input validation
+4. Follow existing test conventions exactly
+5. Run the tests to verify they pass
+6. Commit and push the new tests
+
+## Output
+
+Post a summary comment listing:
+- Tests added (name + what they verify)
+- Coverage areas
+- Any gaps you couldn't cover and why