anatoly-m-maslennikov

Create or refactor portable Codex, Claude, OpenCode, and similar agent skill packages while preserving behavior and safety. Use when the user asks to create, update, refactor, trim, clean up, slim down, make token-cheap, modernize, scriptify, validate, mirror, or reorganize a `SKILL.md` or skill package.

Create or update Obsidian source/reference records for third-party material. Use when saving articles, web posts, site/page collections, and docs pages into 81_REFERENCES as full personal-use local copies with meaningful images downloaded into 82_ASSETS/51_attachments; for websites use a deterministic Python scraper through 51_DEV/10_PROD/obsidian_vault_python. Also use for cloning/downloading repository references into /Users/am/Documents/Repos, placing book/PDF/EPUB references in Google Drive, creating source notes, following `00_META/40_Operations/10_81_REFERENCES authoring rules/00_81_REFERENCES authoring rules HUB.md`, using the reference-title format "author(s) - title (source domain, published)", or cleaning reference-note metadata/templates.

Create persistent Obsidian image asset sets for Colleagues, not guys posts. Use when Codex or Claudian is asked to turn a finished CNG post note into exactly two generated editorial images, a wide post banner and square podcast preview, saved under 82_ASSETS/51_attachments and embedded in a sibling Assets note using the live CNG image prompt chain.

Transform raw Russian Colleagues, not guys podcast transcripts into publishable first-person Obsidian Markdown post drafts in Anatoly/CNG voice. Use when the user provides a CNG transcript or outline and wants a textified-speech article, not a generic summary or retelling.

Generate CNG-style title variants, subtitle variants, and short promo texts from a finished Colleagues, not guys post. Use when the user provides a CNG post and wants current-post-only packaging options without generic marketing tone or old-context leakage.

Maintain Codex custom skill sources in the Obsidian vault and keep per-skill entries under `~/.codex/skills` as symlinks to those vault sources. Use when auditing installed Codex skills, importing real installed skill folders into the vault, relinking broken or copied global skill installs, creating/updating the vault-to-`~/.codex` mirror, or checking that custom skills are symlinked from `52_LLM` rather than copied into `~/.codex`.

Convert Russian HeadHunter or hh.ru resume/CV exports into concise professional resume drafts for a target role. Use when the user supplies a Russian CV, hh.ru PDF/text, or job target and wants a short achievement-focused resume with selected recent experience.

Rewrite an article-generation prompt by comparing source material, an original prompt, and a redacted target article. Use when the user wants generalized style and structure instructions inferred from an edited article without copying lexical replacements.

Turn ambiguous product or domain intent into explicit DSET vocabulary, boundaries, invariants, requirements, scenarios, proof obligations, and decision records. Use before specification acceptance when a feature request has unresolved branches, overloaded terms, hidden actors or states, unclear ownership, conflicting edge cases, or premature solution choices.

Diagnose software or production failures through reproducible evidence, minimization, hypotheses, instrumentation, first-bad-commit analysis, and DSET Back-to-Left provenance. Use when the user asks to investigate, explain, triage, or find the cause of a defect, regression, failed gate, incident, or inconsistent behavior; diagnosis and reporting are read-only unless the user separately authorizes artifact edits or a fix.

Run a bounded disposable experiment to answer a specific design, framework, library, protocol, performance, recovery, or integration question with comparable evidence. Use when uncertainty blocks a DSET Solution Landscape or Decision and production-quality implementation would be premature; never use the prototype as an unreviewed path into production.