jira-start-task

name	jira-start-task
description	Resolve a Jira issue, ensure the correct branch, generate an OpenSpec prompt, and enrich context using Obsidian MCP for documentation and CodeGraph MCP for structural code analysis.
license	Proprietary
metadata	{"author":"codiplab","version":"0.5.0","opencode":{"emoji":"🎯","triggers":["usa el skill jira-start-task","empezar tarea jira","trabajar en issue","generar prompt openspec"],"tags":["jira-cloud","git","workflow","openspec","mcp","obsidian","codegraph","enrichment"],"mcp":{"preferredServer":["jira","codegraph"]}}}

jira-start-task

Starts work on a Jira issue and generates an OpenSpec prompt ready for execution.

---

Template Resolution

Path:

<skill_root>/prompt.md.tpl

Rules:

• MUST exist
• NO fallback allowed

---

Flow

---

1. Resolve Issue

If issueKey is provided

• Fetch via Jira MCP

If NOT provided

• Read codi.toml
• Get issueTracking.projectKey

If missing:

codi.toml not found. Provide issueKey or configure projectKey.

Run JQL:

project = <PROJECT_KEY>
AND assignee = currentUser()
AND statusCategory != Done
ORDER BY updated DESC

---

2. Fetch Issue

Retrieve:

• key
• title
• description

Fail if not found.

3. Parsing

Extract:

• Scenario
• Acceptance Tests
• Sources
• Context Queries
• Code Symbols (via lexical heuristics)

Fallback:

• raw_description

---

4. Content Extraction

Context Queries

Pattern:

mcp obsidian: <query>

---

Code Symbols

The system MUST extract candidate symbols from issue title and description:

Lexical heuristics:

• PascalCase words → class/component names (e.g., "AuthService", "UserController")
• camelCase words → function/variable names (e.g., "getUser", "validateToken")
• Technical terms → "API", "service", "endpoint", "middleware", "handler", "controller", "model", "route", "schema", "mutation", "query"

Fallback:

• empty array [] if no matches

---

5. Data Contract

issue:
  key: string
  title: string

content:
  scenario: string
  acceptance_tests: string[]
  sources: string[]
  context_queries: string[]
  raw_description: string
  extracted_symbols: string[]  # from lexical heuristic extraction

---

codegraph_enrichment: status: "applied" | "partial" | "skipped_no_index" | "skipped_no_symbols" queries: string[] mcp: - tool: "codegraph_explore" | "codegraph_search" | "codegraph_node" | "codegraph_callers" | "codegraph_callees" query: string result: string error?: string

---

6. Enrichment (Obsidian MCP - Smart Context)

This step enriches the prompt with relevant, human-readable context from the Obsidian vault.

---

Execution Rule

Run ONLY if:

• context_queries NOT empty

---

MCP Flow (Per Query)

For each query:

---

1. Search

Primary search tool:

• obsidian_obsidian_simple_search

Optional advanced search (when needed for narrower matching):

• obsidian_obsidian_complex_search

---

2. Selection

The system MUST:

• rank results by textual relevance to the query
• deduplicate by file path (one entry per file)
• select top 3-5 files per query
• require valid selected_path

---

3. Fetch File Content

The system MUST fetch selected files using:

• obsidian_obsidian_get_file_contents (single file), or
• obsidian_obsidian_batch_get_file_contents (multiple files)

with:

• selected_path: <vault_relative_file_path>

---

4. Summarization (CRITICAL)

The system MUST transform fetched note content into a concise summary.

Rules:

• MUST extract only relevant parts for the scenario
• MUST remove noise (excess formatting, unrelated sections)
• MUST be readable plain text
• MUST be 3-6 bullet points
• MUST focus on actionable rules / patterns

---

5. Fallback

If search yields no usable files, or selected paths cannot be fetched:

• list vault files with obsidian_obsidian_list_files_in_vault
• optionally narrow scope with obsidian_obsidian_list_files_in_dir
• retry search/fetch with refined query or narrowed directory scope

---

Enrichment Output

enrichment:
  status: "applied" | "failed" | "skipped_no_context_queries" | "skipped_obsidian_unavailable"
  reason: string
  context_queries: string[]
  mcp:
    - query: string
      tool: "obsidian_obsidian_get_file_contents" | "obsidian_obsidian_batch_get_file_contents"
      selected_path: string
      result: string  # summarized context
      error?: string

---

Rules

• MUST NOT modify original content
• MUST continue on errors
• MUST always return enrichment block

---

7. Enrichment (CodeGraph MCP - Structural Context)

This step enriches the prompt with structural code context from the codebase via CodeGraph MCP queries.

---

Execution Rule

Run ALWAYS — unconditional (unlike Obsidian which requires context_queries).

---

Symbol Extraction

Before querying, extract symbols from the issue title and description using lexical heuristics (see Parsing step). Feed extracted symbols as search queries.

---

MCP Flow

---

1. codegraph_explore (Primary — most information per call)

Use when extracted symbols suggest a bounded code area.

Execution:

• Single symbol: call codegraph_explore with that symbol as the query
• Multiple symbols: call once with up to 3 symbols joined
• Result limit: max 12 files

---

2. codegraph_search

Use when symbol extraction produces names but bounded area is unknown.

Execution:

• Call codegraph_search with symbol name and optional kind filter
• Result limit: top 10 matches
• Follow up with codegraph_node or codegraph_explore if relevant

---

3. codegraph_node

Use when the issue directly references a symbol (file:line format, explicit class/function name).

Execution:

• includeCode: true
• Include full symbol source in output
• Annotate with file path, line range, and callers/callees counts

---

4. codegraph_callers

Use when the issue implies modifying an existing symbol.

Execution:

• Call codegraph_callers for the target symbol
• Limit: 20 callers
• Include in output to show what would break

---

5. codegraph_callees

Use when the issue involves adding new functionality or refactoring.

Execution:

• Call codegraph_callees for the relevant symbol
• Include dependency relations in output

---

Summarization

The system MUST transform raw CodeGraph output into concise summaries:

• Explore results: extract symbol definitions, signatures, key structural relationships
• Node results: include full symbol body in fenced code block with file header
• Error results: capture error message and continue with next tool

---

Fallback

If CodeGraph MCP index returns "not initialized":

• Set status: "skipped_no_index"
• Continue flow without error

If no symbols extracted:

• Set status: "skipped_no_symbols"
• Continue flow without error

If some tools fail and others succeed:

• Set status: "partial"
• Include successful results + error messages

---

8. Render Template

Inputs:

• issue
• content
• enrichment
• codegraph_enrichment

---

9. Output

Ensure output directory exists:

mkdir -p .codi/jira/issues/prompts

Output path:

.codi/jira/issues/prompts/<ISSUE-KEY>.md

---

Validation

Output MUST:

• be valid OpenSpec prompt
• use MUST / SHOULD / MAY

---

Debug

• issue source
• branch
• enrichment status