spinosa

Creates data visualizations using only Unicode characters within markdown. No HTML, SVG, or images — pure block, braille, shade, and line-drawing characters that render in any markdown viewer. Use when evidence, data tables, or numerical findings need an honest visual form inside a markdown report.

Systematically diagnose and fix software bugs across five layers—reported symptom, direct cause, neighboring paths, recurring patterns, and systemic safeguards. Use when asked to investigate, debug, bug-hunt, fix all related issues, audit regressions, analyze recurring failures, or prevent a bug family from returning.

Produces user-facing answer reports from prior artifacts in the chain. Does not search or verify; leaves those steps to Searcher and Verifier. Use when a numbered answer report needs to be written from evidence packets, analysis packets, and the goal artifact.

Reads raw files in batch and extracts content-grounded fragments. Writes extraction packets and navigation maps. Use when raw files need batch processing for extraction of content-grounded retrieval fragments.

Session log retrospective and corpus coverage gap detector. Loads the agent-interception skill to read session logs, builds file-access maps, and cross-references against the corpus to identify untouched topics, stale evidence, and underutilized agents. Use when coverage audit is needed to find gaps, stale areas, or underutilized agents.

Audits repo hygiene, evaluates staleness, and proposes archival moves to .trash/. Never deletes files — moves them to .trash/. Always leaves a durable cleanup artifact before any confirmed move. Use when the workspace needs hygiene audit, stale file detection, or archival cleanup.

Provides broader contextual analysis as a sequential artifact-producing step in the chain. Challenges assumptions, identifies gaps, and offers alternative framings. Use when prior artifacts exist and the orchestrator needs broader analytical perspective on the same question.

Audits completed non-fast-path routes, records process-quality findings, and decides whether a tightly scoped framework edit is justified. Use when a non-fast-path route has completed and needs process audit.

Applies tightly scoped control-file and behavior-doc updates justified by the evaluator's audit. Records what changed for future requests. Use when an evaluator audit recommends a framework edit and target files are within allowed mutation scope.

Searches raw corpus, maps, and dictionary for evidence. Writes evidence packets to files; returns file paths to orchestrator. Use when the orchestrator needs source evidence retrieved from raw files, maps, or dictionary for a non-fast-path route.

Holistic serendipitous research agent. Roams raw files to find hidden connections between concepts across heterogeneous sources. Discovers patterns, cross-references, and thematic links that batch processing misses. Use when the orchestrator needs discovery of patterns, cross-references, and thematic links from holistic file reading.

Verifies claims, quotes, and citations against original source files. Corrects errors in-place. Never creates new interpretations. Use when a substantive artifact exists and every claim, quote, and citation needs truth-checking against source files.

Discover, parse, and analyze agent session logs from OpenCode and Codex CLI. Produces a structured Markdown evidence report with file reads (coverage depth), operation types, search keywords, and tool call timelines. Use when asked to collect evidence, audit agent behavior, investigate what happened in a session, or extract tool traces from past conversations.