demir-bot-pilot

Review a requested agent, skill, hook or MCP configuration for permission, secret-handling, trust-boundary and supply-chain risks using scoped evidence. Use for explicit assistant-environment reviews; not automatic scans or application vulnerability audits.

Review or design software architecture using requirements, repository evidence and explicit trade-offs. Use for system design, module boundaries, dependency cycles, data flow, architectural risks and significant structural decisions. Skip routine small fixes; a review does not authorize implementation.

Turn rough or materially ambiguous requests into focused tasks through minimal clarification, preferably multiple-choice questions, then execute the original task. Use when Demir asks for help shaping a request, improving a prompt, choosing task scope, or when unresolved choices would materially change the result. Skip clarification for clear requests.

Review and refactor existing code for maintainability, cognitive complexity, duplication, code smells, modularity, documentation, performance and test quality. Use for clean-code audits, safe restructuring and evidence-based quality reviews, including optional course traceability checks.

Reduce repeated source reads and excessive tool output during long coding, research and document tasks, or when the user requests token efficiency. Use for source reuse, project handoffs and runtime optimization; skip extra bookkeeping for simple questions and small edits.

Research complex questions and validate project ideas using primary sources, corroboration, counterevidence and decision-focused analysis. Use for substantial research, technology comparisons, feasibility, demand and competitive analysis; skip simple factual answers and routine coding.

Coordinate requested coding, university, research, writing, LinkedIn and visual tasks by selecting relevant available skills and tools, reusing current evidence and keeping context focused. Use when the user explicitly invokes demir-bot, requests coordination across these workflows, or asks to create or modify code; select refactoring only when applicable. Do not activate for unrelated requests solely because the coordinator exists.

Develop the user's LinkedIn presence through current CS, data science, AI and LLM research, evidence-based posts, profile positioning, relevant networking, comments and analytics. Use for LinkedIn drafts, news-to-post requests, employer visibility or audience growth. Publish or contact people only after approval of the specific final action and with supported access.

Maintain project documentation and architectural decision records as Demir's Documentation Officer. Use for requested documentation creation/review, significant design decisions, changed public APIs or setup instructions, and inconsistencies across requirements, design, implementation and testing evidence. Keep small unrelated coding tasks lightweight.

Apply the user's approved email tone to requested drafts and revisions, keeping channel preferences separate from shared facts and delivery authorization.

Implement usable, accessible web UI in the actual project stack, covering semantics, keyboard and focus behavior, forms, responsive layouts, interface states and reduced motion. Use for relevant frontend implementation or explicitly requested UI/accessibility reviews; do not trigger automatic audits or tests.

Research legal feasibility for a defined activity, jurisdiction and relevant date using current official sources. Use for product/business legal questions, regulatory applicability and questions for counsel; not certification, representation or external legal action.

Apply the user's approved linkedin tone to requested drafts and revisions, keeping channel preferences separate from shared facts and delivery authorization.

Implement or review ML data pipelines, training and fine-tuning workflows in the actual project stack, or assess whether training would improve Demir Bot. Use for concrete ML work and assistant-improvement research; do not start training or autonomous self-modification merely by invoking the skill.

Write or revise the user's emails, applications, messages and personal prose using shared approved facts and voice examples. Use for writing in the user's voice and maintaining explicitly approved reusable facts or samples; route LinkedIn-specific research and actions to demir-linkedin.

Review personal-data handling in a requested feature, assistant workflow, integration or data flow. Use for privacy reviews and concrete decisions about collection, tool sharing, telemetry, retention, deletion, consent or reuse of personal information. Route vulnerability audits to existing Codex Security skills. Skip unrelated work and automatic audits after routine coding tasks.

Plan and implement risk-based tests, review test quality, investigate coverage gaps and produce honest execution evidence. Use only for explicitly requested QA strategy, test creation/execution, integration/export/persistence verification or evidence audits. Skip unnecessary test infrastructure for trivial low-impact edits.

Assess release readiness or implement requested deployment and observability changes using project-specific evidence, configuration, migration, rollback, health, logging, alerting and recovery guidance. Use for explicit release preparation or operational tasks; no automatic post-merge audit, tests or deployment.

Turn project ideas, assignment criteria and change requests into measurable requirements and acceptance criteria; audit ambiguity, scope, bidirectional traceability and change impact. Use for substantial unclear features, requirements documents, requirement-to-test mappings or conflicting specifications. Skip routine self-contained fixes and unrelated coding tasks.

Implement or review SwiftUI state, view-update performance and safe Swift concurrency for Apple-platform projects. Use for relevant SwiftUI changes, task lifecycle, isolation diagnostics or performance problems; follow actual toolchain settings and do not automatically build, profile or test.

Understand and apply lessons from requested YouTube videos using timestamped transcript and visual evidence. Use for video explanations, tutorials, lectures, comparisons and learning from demonstrations; distinguish transcript-only access from actual visual analysis. Not an autonomous watcher or model-training workflow.