Skip to main content
GeneralCodename-Inc

spectre-kickoff

👻 | Project kickoff with deep research & MVP pathfinding - primary agent

Stars
142
Source
Codename-Inc/spectre
Updated
2026-05-28
Slug
Codename-Inc--spectre--spectre-kickoff
View on GitHubRaw SKILL.md

// install — copy + paste into any project

mkdir -p .claude/skills && curl -fsSL https://raw.githubusercontent.com/Codename-Inc/spectre/HEAD/plugins/spectre-codex/skills/spectre-kickoff/SKILL.md -o .claude/skills/spectre-kickoff.md

Drops the SKILL.md into .claude/skills/spectre-kickoff.md. Works with Claude Code, Cursor, and any agent that loads SKILL.md files from .claude/skills/.

kickoff

Input Handling

Treat the current command arguments as this workflow's input. When invoked from a slash command, use the forwarded $ARGUMENTS value.

kickoff: Deep Codebase Research + MVP Path Discovery

Comprehensive codebase research, external best practices, and MVP implementation path with file:line evidence. Output: kickoff document with architecture insights, gap analysis, and implementation options saved to docs/tasks/{task_name}/kickoff/.

ARGUMENTS

$ARGUMENTS

Step 1: Acknowledge & Clarify

  • Action — ImmediateReply: Respond before any tools.
    • If ARGUMENTS → acknowledge context, identify: what we're exploring, what decision we're heading toward, what success looks like
    • Else → prompt for project context
    • CRITICAL: No tool calls in this step

Step 2: Gather Context & Decompose

  • Action — ReadMentionedDocs: Read referenced docs FULLY in main context (not subagent). Extract: vision, constraints, decisions made, open questions.
  • Action — DecomposeResearchAreas: Break project into research areas. Consider: components to investigate, directories/files, architectural patterns, data flows, existing code to extend.
  • Action — CreateResearchPlan: Use TodoWrite to track research subtasks.

Step 3: Deep Parallel Research

  • Action — SpawnResearchAgents: Launch parallel agents for comprehensive context.
Agent Task Required Output
@finder Find relevant files, entry points, handlers, models File paths by domain
@analyst Trace data flow, dependencies, behavior, edge cases file:line refs for ALL findings
@patterns Find similar implementations, patterns to follow/avoid Code examples with file:line
@web-research Best practices, similar solutions, pitfalls Findings WITH LINKS

Strategy: locator → analyzer on findings → parallel for breadth. Spawn follow-ups if shallow. Demand file:line evidence.

3rd Party Libs: Use Context7 MCP for central libraries.

  • Action — WaitForAgents: ALL agents must complete before synthesis. Update TodoWrite as each completes.

Step 4: Synthesize Understanding

  • Action — CompileFindings: Synthesize with file:line refs throughout.

    • Codebase: Connect findings across components, document data flow, answer "how does codebase handle [area]?"
    • Architecture: Patterns to follow (with examples), conventions, constraints, reuse vs build new
    • External: Industry patterns (with links), best practices, pitfalls
  • Action — PerformGapAnalysis: Current capabilities (with file refs) vs required capabilities. What's missing vs needs modification.

  • Action — IdentifyMVPAndOptions:

    • MVP: Core value, minimum for value, what to defer (informed by codebase complexity)
    • Options: 2-3 approaches with: summary, key decisions, code to leverage (file refs), new work, effort sense, trade-offs
    • Decision points: Architecture, scope, technology, integration, patterns

Step 5: Generate Document

  • Action — DetermineOutputDir:

    • Derive task_name from context (kebab-case)
    • OUT_DIR = user_specified || docs/tasks/{task_name}/kickoff
    • mkdir -p "$OUT_DIR"
  • Action — CreateDocument: Structure with YAML frontmatter (date, git_commit, branch, repo, topic, tags, status).

    Sections: Title → Metadata → Project Context → Research Summary → Detailed Codebase Findings (by area, file:line refs, code snippets) → Code References (table) → Architecture Insights (patterns, conventions, constraints) → External Research (with links) → Gap Analysis → MVP Suggestion → Implementation Options (2-3 with trade-offs) → Decision Points → Open Questions → Related Resources

  • Action — AddGitHubPermalinks: If on main/pushed, convert file refs to https://github.com/{owner}/{repo}/blob/{commit}/{file}#L{line}

  • Action — SaveDocument: Save as {task_name}_kickoff.md (timestamp variant if exists). CRITICAL: Save before Step 6.

Step 6: Present & Scope

  • Action — PresentSummary:

    Kickoff Complete — saved to {path}

    Vision: [1-2 sentences] What Exists: [key components with file:line refs] Architecture Insights: [patterns, conventions, constraints] External Learnings: [key insights with links] Gap: [have → need] MVP Path: [smallest valuable slice]

    Scoping Questions (before planning):

    1. [Scope/boundary question] — Option A (leverages code:line) vs Option B
    2. [Technical approach] — suggested vs alternative
    3. [Priority/constraint question]
  • Action — EngageInScoping: Wait for response, ask follow-ups. No planning until ambiguities resolved. Research answerable questions instead of asking.

Step 7: Handle Follow-ups

  • Action — HandleFollowUps:
    • If clarifications → update doc with ## Scoping Clarifications [timestamp]
    • If more research needed → spawn agents, demand file:line evidence, update doc
    • Continue until ambiguities resolved

Step 8: Transition to Scope

  • Action — SummarizeAndTransition:

    Ready for Scope Definition

    Established: User problem, user value, key findings (file:line), decisions made, remaining ambiguities

    Options:

    1. "Proceed with scope" → spectre-scope with FROM_KICKOFF=true, SKIP_EXPLORATION=true
    2. "Skip to planning" → spectre-plan with kickoff doc context
    3. "Need to think" → spectre-handoff
  • Action — ExecuteChoice:

    • If scope → invoke spectre-scope with: FROM_KICKOFF=true, KICKOFF_DOC={path}, SKIP_EXPLORATION=true, context summary
    • If planning → suggest spectre-plan
    • If pause → run spectre-handoff