Miosa-osa

Evaluate retrieval and generation quality in RAG pipelines. Separate scoring for retrieval (recall, precision, MRR) and generation (faithfulness, relevance, completeness). End-to-end pipeline assessment with bottleneck identification. Triggers on: "eval rag",…

Design binary pass/fail LLM-as-Judge evaluators. Structured prompt engineering for evaluation: criteria definition, rubric construction, few-shot calibration, and bias mitigation. Produces a ready-to-deploy judge prompt with scoring instructions. Triggers on:…

Language Agent Tree Search - Monte Carlo planning - 92.7% on HumanEval

Self-learning system based on SICA, VIGIL, and Mem0 patterns. Auto-triggers after task completion. Captures patterns, consolidates memory, generates skills, recovers from errors.

Self-improving prompts through meta-level optimization

Optimize token usage through prompt caching and compression

Thought-Action-Observation loop for transparent reasoning

Self-correction via critique loop - 18.5 percentage point improvement

Full-scope penetration testing — reconnaissance, vulnerability discovery, exploitation, post-exploitation, and reporting against authorized targets. Covers network, web app, API, and host pentesting.

Diagnose broken, failing, incorrect, flaky, or unexpectedly slow software. Use whenever a user reports a bug, regression, crash, hang, visual defect, or unexplained behavior, even if they ask directly for a fix.

Build, debug, review, or polish user interfaces with responsive layout, accessibility, and pixel-level visual verification. Use for frontend components, browser UI, terminal UI, resize defects, visual regressions, and interaction problems.

Implement a requested feature or change completely in an existing codebase. Use for multi-file coding tasks, specifications, acceptance criteria, and requests to build or modify behavior rather than merely explain it.

Review code, configuration, permissions, authentication, data handling, and dependencies for exploitable security weaknesses. Use for security reviews, threat analysis, trust boundaries, secrets, authorization, sandboxing, and dangerous command handling.

Create or improve reusable OSA skills with progressive disclosure, precise selection descriptions, supporting resources, and realistic evaluations. Use when users ask to create, package, test, optimize, or maintain a skill or skill library.

Build features and bug fixes test first using a strict red, green, refactor loop. Use for TDD requests, regression tests, behavior changes with clear acceptance criteria, and risky logic that benefits from executable specification.

Recover the work a fleet wave dropped when two parallel nodes edited the same file — read each claimant's version out of its worktree, reconcile them into one correct file, and re-run the authoritative gate instead of re-running the wave.

Optimal Engine is a self-hosted second brain and operating engine for human and AI workspaces. It preserves messy sources, organizes them into workspaces and Nodes, classifies Signals, turns extracted assertions into reviewable Claims, promotes accepted…

Analyze productivity trends, explain metrics, and identify bottlenecks. Answer questions like "why are tasks overdue" or "how am I doing this week". Provide actionable insights based on workspace data.

Create and configure custom dashboards with widgets. Add task summaries, burndown charts, project progress, upcoming deadlines, and metric cards. Group tasks by status, project, or priority. Set default dashboards.

Configure notification preferences, set quiet hours, and manage alerts. Control how and when you receive notifications for tasks, mentions, deadlines, and team activity.

Create and manage projects. Add team members, set deadlines, track status, and organize work. Configure project settings, templates, and workflows.

Create, update, complete, and delete tasks. Filter by status, project, priority, or due date. Bulk operations for efficiency. Assign tasks to team members and manage task dependencies.

Use when writing browser QA findings.

Use when turning raw support context into a precise triage note.

Use before production deploys.

Use for diagnosing failing CI.

Use for reviewing code changes.