interfluence: Analyze Writing & Generate Voice Profile

You are analyzing the user's writing corpus to generate comprehensive voice profiles.

Process

1. Load the corpus using corpus_get_all to get all writing samples

2. Load existing profile(s) (if any) using profile_list then profile_get — you'll refine rather than replace

3. Load accumulated learnings (if any) using learnings_get_raw — incorporate these

4. Delegate to the voice-analyzer agent using the Task tool:

   • Launch the interfluence:voice-analyzer agent (subagent_type)
   • The agent performs comparative analysis: classifies samples by context, extracts cross-context invariants as the base profile, and generates per-context deltas
   • The agent presents classification summary before generating profiles
   • The agent saves profiles via profile_save (base) and profile_save(projectDir, content, voice) (deltas)

5. After analysis completes, show the user:

   • Available voices using profile_list
   • Summary of what was generated: "Created base profile + 2 voice deltas (blog, docs)"
   • The most distinctive traits from each voice
   • If any samples were unclassified, note this

Existing Profile Handling

• If the user already has profiles, the analyzer should merge new insights rather than replacing — preserve user refinements
• If this is the first multi-voice analysis on an existing single profile, the analyzer will restructure: existing profile becomes the base, new deltas are generated from comparative analysis

Important

• The voice profile must be specific enough to actually follow, not vague platitudes
• Quote the author's actual words as examples wherever possible
• If there are too few samples for confident analysis, note the uncertainty and suggest adding more
• Classification happens in-memory during analysis — tags are NOT persisted to corpus-index.yaml