
from crushable-wingman-skill37
A persistent texting and DM coach that drafts replies, extracts chat screenshots, and remembers profiles so it never re-asks the same questions.
Provides a persistent dating wingman assistant that turns messy chat context into actionable reply suggestions, date ideas, and relationship health analysis. The skill defines a structured workflow: load user and crush profiles, extract transcripts from screenshots (with confirmation before coaching), ask targeted clarifying questions only when context is missing, and deliver ranked reply options with follow-up branches. Profile snapshots and log entries are written after each session to maintain continuity.
Activate when a user asks for message rewrites, DM openers, reply suggestions, chat screenshot analysis, or date ideas — especially in multi-turn coaching scenarios where persistent state (profiles, logs) improves advice quality over time. Also handles emotional check-ins by acknowledging feelings before suggesting next steps.
~/.codex/state/crushable-wingman/.Agents with long-term context and local filesystem access: Claude/Claude Code, Codex, and self-hosted assistants able to read/write local Markdown state files.
Crushable Wingman is a dating/texting coach skill that persists user and crush profiles plus conversation logs in local Markdown files. It includes a well-structured Python CLI (wingman_store.py) for state management — CRUD for profiles, memory snapshots with enforced limits, and append-only logs. The script runs cleanly for most commands (paths, list, validate, missing), but the init command requires external template files from a references/ directory that aren't included in the script bundle, so it fails standalone. The SKILL.md is thorough and well-organized with clear triggers, workflow steps, safety boundaries, and progressive disclosure via references.
No security concerns — the script does no network calls, uses no credentials, and only writes to local Markdown files. The skill explicitly includes safety boundaries against harassment/coercion. Niche use case (dating coach) limits broad usefulness, but it's well-crafted for its audience. The main script is ~550 lines of clean, well-structured Python with argparse subcommands, proper error handling, and data validation (snapshot limits). Architecture follows the skill spec well: frontmatter complete, references/ directory, scripts/ directory, clear output contracts.