Fix pipeline friction: permissions, branch setup, skill file access

plugins/dev-team/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "dev-team",
-  "version": "1.2.2",
+  "version": "1.2.3",
   "description": "Dev-team agent pipeline: researcher, developer, reviewer, and debugger agents for implementing Jira tasks and fixing GitHub issues.",
   "commands": "./commands"
 }

plugins/dev-team/agents/task-runner.md

@@ -27,7 +27,8 @@ Parse the following fields from your prompt:
 
 - `agent` — sub-agent type to spawn via the `Agent` tool (used for display/logging in
   the context header only; does not affect routing or tool selection)
-- `skill` — name of the skill the sub-agent should invoke
+- `skill` — name of the skill the sub-agent should execute
+- `plugin_root` — absolute path to the dev-team plugin directory (used to load skill files)
 - `context_file` — absolute path to the pipeline context file
 - `args` — (optional) positional arguments to present to the skill
 - `read_sections` — comma-separated list of section names to read from the context file
@@ -40,30 +41,32 @@ Use the `Read` tool to read `context_file`. Extract the content of each section
 `read_sections`. A section begins at `<!-- section:Name -->` and ends at the next
 `<!-- section:` sentinel or end of file.
 
-### Step 2 — Present context
+### Step 2 — Load and substitute skill content
 
-Present the extracted sections as a quoted block in your conversation before invoking
-the skill. Format:
+Using the `Read` tool, read: `<plugin_root>/commands/<skill>.md`
 
-```
-## Context from pipeline
+Strip the YAML frontmatter — everything from the opening `---` line to the closing `---`
+line (inclusive), plus any blank line immediately after.
 
-### <Section Name>
-<content>
-```
+Apply the following substitutions to the remaining content using the values already
+extracted from the context file:
 
-Also present the `args` value (if provided) as:
-```
-## Arguments
-<args>
-```
+| Variable | Source |
+|----------|--------|
+| `$ARGUMENTS` | `args` field from this prompt |
+| `$TASK_BRIEF` | "Researcher Brief" section content |
+| `$WORK_SUMMARIES` | "Implementation Summary" section + all "Fix N" sections, concatenated with `\n\n---\n\n` |
+| `$ISSUES` | "Last Failure" section content (or "Validate Result" if "Last Failure" is empty) |
+| `$PR_URL` | "PR URL" section content |
+| `$CONTEXT_FILE` | `context_file` path |
 
-Set `$CONTEXT_FILE` to the value of `context_file` — skills that need mid-task reads
-use this substitution.
+Leave any variable with no matching value unchanged.
 
 ### Step 3 — Spawn the sub-agent
 
-Spawn a sub-agent using the `Agent` tool:
+Spawn a sub-agent using the `Agent` tool, passing the substituted skill content directly
+as `## Instructions`. The sub-agent executes the instructions without needing to load any
+skill files.
 
 ```
 Agent(
@@ -77,13 +80,13 @@ Agent(
 ## Arguments
 <args>
 
-## Skill
-<skill>
+## Instructions
+<substituted skill content>
 """
 )
 ```
 
-The sub-agent invokes the named skill and returns its full output. Capture that output
+The sub-agent executes the instructions and returns its full output. Capture that output
 as the skill result for Step 4.
 
 ### Step 4 — Write result to context file (Edit only — never Write)

plugins/dev-team/agents/workspace-setup.md

@@ -0,0 +1,15 @@
+---
+name: workspace-setup
+description: >
+  Sets up the git workspace before a pipeline run. Finds the epic branch from
+  the spec or Jira, fetches latest, and creates the task branch.
+model: haiku
+tools:
+  - Read
+  - Bash
+  - mcp__08e9ccd3-4093-4425-adec-d98ea766a759__getJiraIssue
+---
+
+You are the workspace-setup agent for the dev-team pipeline.
+Your only job is to set up the correct git branch before implementation begins.
+Execute the instructions in `## Instructions` in your prompt.

plugins/dev-team/commands/create-branch.md

@@ -1,31 +1,25 @@
 ---
-description: Ensure the current work runs on a feature branch derived from a work-item ID and title/brief text.
-argument-hint: <work-item-id> "<issue-title-or-task-brief-sentence>"
+description: Ensure the current work runs on a dev/claude/<work-item-id> branch.
+argument-hint: <work-item-id>
 user-invocable: false
 ---
 
 ## Inputs
 
-- Work item ID: first token in `$ARGUMENTS` (e.g. `Issue-444`, `ADR-172`)
-- Slug source text: remaining argument text (quoted when it includes spaces)
+- Work item ID: `$ARGUMENTS` (e.g. `Issue-444`, `ADR-172`)
 
-If either input is missing, stop and print:
+If missing, stop and print:
 
-> Usage: `/create-branch <work-item-id> "<issue-title-or-task-brief-sentence>"`
+> Usage: `/create-branch <work-item-id>`
 
 ## Steps
 
 1. Check the current branch:
    ```bash
    git branch --show-current
    ```
-2. If already on `dev/claude/<work-item-id>-*`, stop (nothing to do).
-3. Derive a slug from the source text:
-   - lowercase
-   - replace spaces and underscores with hyphens
-   - remove non-alphanumeric/non-hyphen characters
-   - trim to 40 characters
-4. Create and switch:
+2. If already on `dev/claude/<work-item-id>`, stop (nothing to do).
+3. Create and switch:
    ```bash
-   git checkout -b dev/claude/<work-item-id>-<slug>
+   git checkout -b dev/claude/<work-item-id>
    ```

plugins/dev-team/commands/dev-team.md

@@ -118,6 +118,7 @@ results = await [
   Agent(subagent_type="task-runner", prompt="""
 agent: <item.agent>
 skill: <item.skill>
+plugin_root: <CLAUDE_PLUGIN_ROOT>
 context_file: <item.context_file>
 args: <item.args or "">
 read_sections: <item.read_sections joined by ", ">

plugins/dev-team/commands/developer-implement.md

@@ -16,21 +16,21 @@ $TASK_BRIEF
 
 ## Steps
 
-### 0 — Ensure feature branch
+### 0 — Ensure task branch
 
 Check the current branch:
 
 ```bash
 git branch --show-current
 ```
 
-If the branch name contains `$ARGUMENTS` (the work item ID, e.g. `Issue-123`), the correct
-branch is already active — proceed to Step 1.
+If the branch is `dev/claude/$ARGUMENTS`, proceed to Step 1 — the workspace is already
+set up.
 
-Otherwise, invoke the shared branch-creation skill:
+Otherwise, create and switch to the task branch:
 
 ```bash
-/create-branch $ARGUMENTS "<task brief first sentence>"
+git checkout -b dev/claude/$ARGUMENTS
 ```
 
 Do not push — the pipeline pushes after validation passes.

plugins/dev-team/commands/workspace-setup.md

@@ -0,0 +1,68 @@
+---
+description: Find the epic base branch and create the task branch for a pipeline run.
+argument-hint: <work-item-id> <spec-path>
+user-invocable: false
+---
+
+## Inputs
+
+Work item ID: first token of `$ARGUMENTS`
+Spec path: second token of `$ARGUMENTS`
+
+## Steps
+
+### 1 — Check if already set up
+
+```bash
+git branch --show-current
+```
+
+If already on `dev/claude/<work-item-id>`, output `setup_done` and stop.
+
+### 2 — Find the epic ID
+
+Read the spec file at the spec path. Look for an Epic reference — a Jira key pattern
+(e.g. `ADR-200`) that appears in headings like `Epic:`, `Parent:`, or in the title block.
+
+If not found in the spec, query Jira:
+- Call `mcp__08e9ccd3-4093-4425-adec-d98ea766a759__getJiraIssue` with the work item ID
+- Look for a `parent` or `epic` field on the issue
+
+If no epic is found, proceed with `main` as the base branch.
+
+### 3 — Find the base branch
+
+If an epic ID was found (e.g. `ADR-200`):
+
+```bash
+git fetch origin
+git branch -r | grep "feature/<epic-id>"
+```
+
+Use the first matching branch (e.g. `origin/feature/ADR-200-infrastructure`), stripping
+the `origin/` prefix. If no match is found, fall back to `main`.
+
+If no epic was found, check for the nearest `feature/*` ancestor of the current HEAD:
+
+```bash
+git branch -r --merged HEAD | grep "feature/"
+```
+
+Use the closest ancestor `feature/*` branch. If none, use `main`.
+
+### 4 — Check out base branch and pull
+
+```bash
+git checkout <base-branch>
+git pull origin <base-branch>
+```
+
+### 5 — Create task branch
+
+```bash
+git checkout -b dev/claude/<work-item-id>
+```
+
+### 6 — Output
+
+Output exactly: `setup_done`

plugins/dev-team/scripts/dev_team.py

@@ -199,6 +199,7 @@ class PipelineContext:
     build_log: str = ""
     test_log: str = ""
     debug_report: str = ""
+    workspace_setup: str = ""
     signoff_review: str = ""
     signoff_research: str = ""
     signoff_build_result: str = ""
@@ -239,6 +240,9 @@ def save(self, path: Path) -> None:
             f"# {self.work_item_id} Dev Team Context",
         ]
 
+        if self.workspace_setup:
+            lines += ["", "<!-- section:Workspace Setup -->", "", self.workspace_setup.strip()]
+
         if self.debug_report:
             lines += ["", "<!-- section:Debug Report -->", "", self.debug_report.strip()]
 
@@ -308,6 +312,7 @@ def load(cls, path: Path) -> "PipelineContext":
             pass
 
         sections = _parse_sections(body)
+        ctx.workspace_setup = sections.get("Workspace Setup", "")
         ctx.debug_report = sections.get("Debug Report", "")
         ctx.brief = sections.get("Researcher Brief", "")
 
@@ -502,6 +507,43 @@ def handle_results(self) -> str:
         ...
 
 
+class SetupWorkspaceStep(Step):
+    handles = "setting-up"
+    _PENDING_KEY = "setup"
+
+    def __init__(self, ctx: "PipelineContext", context_path: Path) -> None:
+        self._ctx = ctx
+        self._context_path = context_path
+
+    def get_actions(self) -> list[dict]:
+        ctx = self._ctx
+        if ctx.workspace_setup:
+            return []
+        return [{
+            "action": "spawn_agent",
+            "message": f"Setting up workspace branch for {ctx.work_item_id}.",
+            "agent": "workspace-setup",
+            "skill": "workspace-setup",
+            "args": f"{ctx.work_item_id} {ctx.spec_path}",
+            "context_file": str(self._context_path),
+            "read_sections": [],
+            "write_section": "Workspace Setup",
+            "result_format": "setup_done | failed",
+        }]
+
+    def handle_results(self) -> str:
+        ctx = self._ctx
+        if ctx.workspace_setup:
+            _handle_agent_success(ctx)
+            return "setup_done"
+        _handle_agent_failure(ctx)
+        _check_and_trigger_troubleshooter(
+            "consecutive_failures", CONSECUTIVE_FAILURES_THRESHOLD,
+            ctx.consecutive_failures, ctx, self._context_path,
+        )
+        return "failed"
+
+
 class FindSpecStep(Step):
     handles = "spec-finding"
 
@@ -1125,6 +1167,7 @@ def __init__(
         self.machine = StateMachine(workflow.transitions, initial=ctx.state)
         self.step_handlers: dict[str, Step] = {
             "spec-finding": FindSpecStep(ctx),
+            "setting-up": SetupWorkspaceStep(ctx, context_path),
             "debugging": DebugStep(ctx, context_path),
             "researching": ResearchStep(research_skill, ctx, context_path),
             "implementing": ImplementStep(ctx, context_path),

plugins/dev-team/scripts/implement-task-plan.md

@@ -2,7 +2,8 @@
 stateDiagram-v2
     [*] --> init
     init --> spec-finding : start
-    spec-finding --> researching : spec_found
+    spec-finding --> setting-up : spec_found
+    setting-up --> researching : setup_done
     researching --> implementing : research_done
     implementing --> validating : impl_done
     validating --> fixing : build_failed