roktas/ajans

Bash and POSIX shell scripting judgment for reusable scripts, automation, CLI wrappers, dotfiles, and CI: interpreter boundaries, status propagation, quoting, globs, temporary resources, portability, style, and validation. Use only when producing or materially revising substantive Bash or POSIX shell code. Do not activate for routine technical Q&A, repository browsing, read-only inspection, tool use, one-off commands, file types, fenced examples, or incidental shell mentions.

C implementation judgment for language-standard and platform boundaries, public headers and ABI, ownership, allocation, lifetime, integer behavior, undefined behavior, threading, native builds, and C tooling. Use only when producing or materially revising substantive C code. Do not activate for routine technical Q&A, repository browsing, read-only inspection, compiler or tool use, file types, fenced examples, project labels, or incidental C mentions.

Targeted review or removal of formulaic generative or AI-like writing artifacts while preserving meaning and voice. Use only when the user explicitly invokes deai, explicitly asks for de-AI, de-slop, humanization, or formulaic- artifact cleanup, or a more-specific project workflow explicitly requires such a pass. Do not activate automatically for ordinary drafting, revision, translation, or because prose merely appears polished or machine-written.

Go implementation guidance for modern language and library idioms plus local Go conventions. Use only when producing or materially revising substantive Go code. Do not activate for routine technical Q&A, repository browsing, read-only inspection, Go tool use, file types, fenced examples, project labels, or incidental Go mentions.

Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, uses any "grill" trigger phrase, or says "beni sorgula", "detaylı tartışalım", or "ayrıntılı tartışalım".

Naming judgment for durable names across code, files and directories, paths, commands, skills, modules and packages, types, functions, variables and fields, APIs, configuration keys, assets, and named concepts. Use whenever creating, renaming, or proposing such a name, even when naming is incidental to a larger task. Do not activate for ordinary prose wording or read-only inspection of existing names.

Ruby implementation judgment for modern syntax, models, style, public APIs, gems and loading, Bundler, tests, typing, visibility, and YARD. Use only when producing or materially revising substantive Ruby code. Do not activate for routine technical Q&A, repository browsing, read-only inspection, Ruby tool use, file types, fenced examples, project labels, or incidental Ruby mentions.

Test design for behavior, public seams, regression coverage, integration tests, and test-first development. Use when writing or materially revising tests, reviewing test design, choosing test seams, or when tests are a substantive part of a bug fix or feature. Do not activate merely to run an existing test command or for routine verification. Apply the TDD loop only when the user explicitly requests TDD, test-first work, or red-green-refactor.

TeX-family judgment for source and generated artifacts, engines, document classes, packages, macros, bibliography, build convergence, diagnostics, and rendered output. Use only when producing or materially revising substantive TeX-family source whose TeX behavior matters. Do not activate for routine technical Q&A, repository browsing, read-only inspection, compiler or tool use, file types, fenced examples, rendered PDFs, project labels, or incidental TeX mentions.