Impersonator

Impersonator Skill

Purpose

Generate a simulated CEET artifact pack for any existing ceet-<role> folder without running a first-person interview.

This skill supports two modes:

1. Public Figure Mode — infer from public, well-documented sources.
2. Repository Author Mode — infer from an author's commit/code history in a repository the operator can access.

Inputs

Required:

1. target_role — one of the existing ceet-* folders.
2. mode — public_figure or repo_author.
3. output_slug — short kebab-case identifier for the output folder (e.g., backend-netflix-tech-blog).

Mode-specific:

1. public_figure (for public_figure mode), optional focus_context.
2. repo, author (for repo_author mode), optional since/until.

Non-negotiable output label

Every generated file MUST include this banner near the top:

Simulation Notice: This profile is inferred from external evidence, not from a first-person CEET interview. Treat as a draft hypothesis, not ground truth.

Run order

1. Validate role. Confirm target_role exists and contains templates/.
2. Load directive structure. Parse ceet-<role>/templates/cognitive-profile.md to detect expected {directives.domain.field} keys and enumerate every template file under ceet-<role>/templates/ (agents, skills, commands, rules, hooks).
3. Collect evidence.
   • public_figure: documented speeches, interviews, writing, public decisions.
   • repo_author: commit messages, diffs, authored files, review footprint (if available).
4. Build evidence matrix. For each directive field, map evidence + confidence (high|medium|low).
5. Fill directives carefully.
   • If confidence is low or evidence is sparse, write unknown and explain why.
   • Do not fabricate precision.
6. Generate full draft pack matching the target role artifact shape:
   • cognitive-profile.md
   • agents/*
   • skills/*
   • commands/*
   • rules/*
   • hooks/* (if role has hooks)
7. Write output under the top-level examples/ready-to-use/ folder, not inside the canonical role templates:
   • examples/ready-to-use/<output-slug>/
8. Attach provenance artifacts in the output folder:
   • evidence-map.md listing every populated directive with source rationale and confidence.
   • README.md explaining what's in the pack and how to copy it into Claude / ChatGPT / Cursor / Copilot / etc.

Mapping rules

1. Map evidence to behavioral directives, not biographical trivia.
2. Prefer repeated patterns over one-off anecdotes.
3. Use direct quotes only when verifiable.
4. Keep role fit explicit: if evidence does not fit the role context, mark it as uncertain.

Refusal and safety rules

1. Refuse requests to present generated output as verified identity or endorsement.
2. Refuse private-person profiling or doxxing-like tasks.
3. For repository mode, proceed only when operator confirms authorization.
4. Avoid sensitive trait inference (medical, political, religion, sexuality, etc.) unless explicitly and publicly work-relevant and necessary.
5. Never claim certainty where evidence does not support it.

Completion criteria

A run is complete only when:

1. Full target-role artifact set is generated in examples/ready-to-use/<output-slug>/.

2. Every file contains the Simulation Notice.

3. evidence-map.md exists and covers all populated directive fields.

4. README.md documents the pack and how to use it in any AI tool.

5. Unknown/low-confidence areas are explicit.

6. The validator passes:

   python3 impersonator/scripts/validate_pack.py \
     examples/ready-to-use/<output-slug>
   

   This script enforces criteria 2–5 deterministically. It checks the Simulation Notice banner, a minimum of evidence rows, the absence of unfilled {directives.*} placeholders, and that every directive cited in cognitive-profile.md also appears in evidence-map.md. A non-zero exit code means the pack must not be published.