athena-work（雅典娜工作）

Athena Brief Processor

Process .athenabrief research packages exported from the Athena note-taking app. These are ZIP archives containing curated notes, AI-generated summaries, a research brief, and binary assets arranged for progressive disclosure.

Package Structure Overview

package.athenabrief (ZIP)
├── instructions.md        # Static reading guide (start here)
├── brief.md               # Research brief, report, and note index
├── summaries.json         # Machine-readable note summaries
├── manifest.json          # Full machine-readable index and statistics
├── references/*.md        # Full note content with YAML frontmatter
└── assets/*               # Binary attachments (images, PDFs)

Read references/athenabrief-spec.md for full schema details when encountering unusual fields or edge cases.

Processing Workflow

Step 1: Extract and Orient

Extract the .athenabrief ZIP to a temporary working directory. Read instructions.md first to understand the package's progressive disclosure model. This file is static across all packages and confirms the reading order.

Step 2: Progressive Disclosure Reads

Read files in this order, stopping as early as possible:

1. brief.md — Read the research objective, suggested web research, research report, and note index table. This provides the high-level context and tells you what the brief is about.
2. summaries.json — Read the machine-readable note summaries. Each entry has generalSummary, contextualSummary, keyContributions, and knowledgeEntries. This gives you enough detail to plan your approach.
3. Short-circuit check — Decide whether you need deeper detail (see decision matrix below). In ~80% of cases, brief.md + summaries.json provide sufficient context.
4. manifest.json — Read only when you need statistics, relevance scores, sub-query coverage, or the full research plan. Skip if summaries.json was sufficient.
5. **references/*.md** — Read individual notes only when you need the full text. Use the manifest or summaries to identify which notes are relevant. Each note has YAML frontmatter with metadata and the full markdown body.
6. **assets/*** — Read binary assets only when referenced in the work or explicitly requested by the user.

Step 3: Short-Circuit Decision Matrix

After reading brief.md and summaries.json, use this matrix:

Step 4: Engagement Rules

Ask the user for clarification when ANY of these apply:

• Imprecise brief — The research objective is vague and multiple interpretations exist
• Fuzzy search instructions — The web research instructions are ambiguous
• Insufficient knowledge — The brief references domains you cannot reliably address
• Contradictory data — Notes contain conflicting information with no clear resolution
• Scope ambiguity — Unclear whether the user wants a summary, analysis, new content, or something else

When the brief contains clear instructions and no ambiguity exists, proceed without asking. Zero-instruction processing is a core feature: the brief itself is the instruction set.

Step 5: Execute and Package

Execute the work based on the brief content and any user instructions. Delegate to other installed skills as needed (e.g., web search, image generation, document creation).

When producing notes as output, ALWAYS use this mandatory header format:

# note title

> short description

---

note content here...

The # title is the note's display name. The > description is a one-line summary. The --- separator always precedes body content.

After completing the work, ALWAYS use the athena-package skill to package results for import back into Athena. Pass all produced notes, relevant assets, and appropriate aurora tags to the packaging step.

Metadata Precedence

When the same data appears in multiple files:

• manifest.json is authoritative for IDs, paths, and scores
• summaries.json is authoritative for note summaries
• YAML frontmatter in references is convenience — use when reading individual notes
• brief.md tables are human-readable summaries — use for quick orientation

Multi-Brief Workflows

When the user uploads multiple .athenabrief packages:

1. Process each brief independently first
2. Look for cross-brief connections using knowledgeEntries (shared entities, themes, linked notes)
3. Synthesize findings across briefs when the user asks for combined analysis
4. Package all results into a single .athena package unless the user requests separate packages

Delegation Patterns

The brief may require capabilities beyond text processing:

Always check what skills and tools are available before promising delegation.

Error Handling

Graceful Degradation Hierarchy

If files are missing or malformed, fall back in this order:

1. brief.md + summaries.json (ideal path)
2. brief.md + manifest.json (summaries unavailable)
3. brief.md alone (both JSON files broken)
4. manifest.json + references/*.md (brief missing)
5. references/*.md scanned directly (minimal package)

At each level, warn the user about missing data and reduced accuracy.

Example

Input: User uploads q1-strategy.athenabrief

Processing:

1. Extract ZIP, read instructions.md (standard reading guide)
2. Read brief.md — Research objective: "Review Q1 product strategy and roadmap priorities"
3. Read summaries.json — 5 notes covering roadmap, standups, competitor analysis, sprint retros, launch checklist
4. Short-circuit: User said "summarize this brief" — brief.md + summaries.json sufficient
5. Produce executive summary note + key findings note + action items note

Output: Hand off 3 notes to athena-package skill:

• "Q1 Strategy Summary" (aurora: focus)
• "Key Findings" (aurora: explore)
• "Action Items" (aurora: commitments)

Reference Materials

• references/athenabrief-spec.md — Full .athenabrief format specification: ZIP structure, brief.md sections, summaries.json schema, manifest.json schema, reference note YAML frontmatter, asset naming conventions. Read when encountering unusual fields, debugging malformed packages, or building custom extraction logic.
• references/processing-strategies.md — Edge case handling, token optimization strategies, graceful degradation details, and advanced multi-brief workflows. Read when facing complex briefs, performance concerns, or unusual package configurations.