davidbirchall/squad-skills-knowledge

Making and recording decisions well - clarifying the decision and criteria, generating options, classifying reversible (two-way door) vs irreversible (one-way door) choices, weighing options with a decision matrix, running a pre-mortem, and writing a decision record with context, options, decision, and consequences. Load when choosing between alternatives, committing to a direction, or documenting why a choice was made.

Writing accurate, durable documentation that the user and future AI sessions can rely on, using the Diataxis framework - tutorials, how-to guides, reference, and explanation - plus structure, plain language, verifiability, and maintenance. Load when creating or restructuring guides, READMEs, procedures, runbooks, or reports, or when existing documents mix purposes and are hard to use.

Capturing and organizing knowledge as durable, atomic, linked notes - fleeting vs literature vs permanent notes (Zettelkasten), one idea per note, source attribution, linking, tagging, and periodic review. Load when turning research, reading, or experience into reusable notes, or when a note collection is hard to search or has gone stale.

Structuring a durable plain-Markdown personal knowledge base that one user and AI sessions can both query - folder layout (inbox, notes by topic, decisions, references, glossary), note anatomy (title, summary line, date, source, status, links), naming conventions, atomic notes versus long documents, linking, index and map-of-content files, minimal tagging, retrieval-friendly writing (front-loaded summaries, stable headings, short files), maintenance (prune, merge, mark stale, last-verified dates), repo notes versus the knowledge base, keeping secrets out, and git versioning. Load when creating, reorganizing, searching, or maintaining a personal notes folder, or deciding where a piece of knowledge should live.

Getting competent in an unfamiliar domain quickly - defining the target capability and a question list, finding the 20 percent core in primary sources and official docs, building a glossary first, drawing a mental model of the domain, learning through small runnable experiments, comparing two or three sources and flagging disagreement, calibrating confidence and marking unverified claims, keeping a learning log, applying a stopping rule, turning learnings into notes or a skill, and knowing when to ask the user for context only they hold. Load when starting work in a domain, tool, or technology you do not yet know well, or when asked to get up to speed on a subject before acting.

Structured research for any topic - framing a research question, planning searches, evaluating and triangulating sources, assigning confidence levels, citing sources, and writing up findings with open gaps. Load when asked to investigate, compare options, verify a claim, or gather evidence before capturing notes or making a decision.

Reviewing any written work, research, plan, code, or decision with a checklist and producing actionable findings - agreeing scope and criteria, separating blocking from non-blocking issues, tying findings to evidence, and re-checking fixes. Works as self-review or an AI fresh-eyes pass after a cooling-off gap. Load when asked to review, critique, proofread, fact-check, or check a document, finding, or decision record.

Summarising documents, articles, transcripts, code, and discussion threads into reusable notes - purpose-first framing, a fixed structure of claims, evidence, numbers, caveats and open questions, keeping faithful and interpretive layers separate, exact quotes with citations (source, location, date), compression ratios, chunking and merging long sources, contradiction flags, no invented details, a verification pass against the source, output templates (brief, detailed, comparison table), and staleness dating. Load when asked to summarise, digest, extract, or compare sources, or when turning raw material into a note that later work will rely on.