ai-skills

Guides the user through designing and creating the complete memory architecture for an AI agent OR for a workspace — MEMORY.md index, tiered topical store, update triggers, retention policies, connectivity test, setup report. Supports THREE Step-0 modes — (1) workspace memory; (2) per-agent memory (requires IDENTITY.md / SOUL.md / RULES.md); (3) workspace-confirm-and-update for boilerplate-derived projects (validates an existing MEMORY.md + memories/ tree structurally, additively merges anything missing without overwriting pre-existing content, surfaces every conflict for user decision, then runs Steps 8–10 unchanged). Use to build, design, or finalise a MEMORY.md / memory architecture / memory backend. Also fires as Phase 4 of build-agent and as the body of configure-workspace-memories during boilerplate bootstrap.

Reorganizes an agent's memory/daily/ folder into a progressively nested hierarchy of monthly (YYYY-MM/) and quarterly (YYYY-QN/) sub-folders, each anchored by a synthesized summary README.md. Keeps the memory/daily/README.md entry point small and navigable regardless of how many session logs have accumulated. Trigger when: flat daily files for a completed calendar month are still at memory/daily/ root; monthly folders for a completed quarter are not yet folded into a YYYY-QN/ folder; the user says "reorganize daily logs", "archive last month's logs", "compact this agent's daily memory", or "reorganize daily logs for [agent-name]". Scoped to exactly one agent per run — always ask which agent if ambiguous. Hard dependency on format-md-for-progressive-disclosure. Does not touch MEMORY.md, topical stores, or any other agent's files.

Weekly memory maintenance orchestrator for a single agent. Audits all eight Tier 2 memory stores, detects cross-store contradictions, delegates daily-log reorganization to compact-agent-daily-memories and topical-store compaction to compact-agent-topical-memories, rebuilds the MEMORY.md index (enforcing the 4 KB limit), and produces a unified maintenance report. Use this skill — not the sub-skills directly — when full weekly maintenance is needed: MEMORY.md index approaching 4 KB, both daily and topical stores need attention, or the user says "run memory maintenance for [agent-name]", "compact this agent's memory", or "run weekly maintenance". Scoped to one agent per invocation.

Compacts the seven Tier 2 topical memory stores (decisions/, projects/, activities/, lessons/, preferences/, people/, analysis/) for one agent. Merges related entries, archives lifecycle-complete records, and prunes expired content — preserving all decision reasoning and never resolving contradictions autonomously. Use when: a topical store README.md exceeds 5 KB; a store sub-folder has 10+ files eligible for merge or archive; the user says "compact decisions", "archive completed projects", "merge lessons", "clean up the projects", or "reorganize memory stores". Also triggered by diagnostic phrases ("are there contradictions?", "show stale entries", "which projects are not archived?") — diagnostic mode audits without writing anything. Invoked by skill-compact-memories for weekly memory maintenance. Requires format-md-for-progressive-disclosure sub-skill. Scoped to one agent per run — always asks which agent if ambiguous. Does not touch memory/daily/.

Adds or updates MAGI-compliant YAML front matter in markdown files so AI agents and indexing tools can discover, classify, and retrieve documents without reading the full body. Use this skill whenever the user wants to add front matter to a markdown file, make a markdown file discoverable by AI agents, tag or describe a file for agentic retrieval, prepare documents for progressive discovery, or update existing front matter metadata. Also trigger when the user mentions "front matter", "MAGI", "progressive discovery metadata", or asks to enrich a .md file with structured metadata — even if they don't use those exact terms.

Captures pending action items into uniform, ID-addressable, durable records — one markdown file per action plus a shared INDEX.md — so they survive past the originating session and remain queryable across agents and humans. Use when the user types /record-action or /action-needed; says "record this action", "add this to the action list", "track this as a todo", "capture this as an action item", "we need to", "someone should", "action item:", "TODO:", or "follow-up:"; or when an agent such as chat-session-scribe invokes it programmatically at session-end. A MEMORY.md path is a hard prerequisite — the skill stops and asks for it if not provided. Performs ASK-BACK for missing owner and due-date. Auto-generates a UUIDv7 ID (time-sortable). Dual-writes to shared workspace memories (source of truth) and, when the caller is an FBO-framework agent, a local copy in the agent's own memories. Companion skill: record-action-taken resolves actions by ID.

Resolves a previously-recorded pending action by marking it done, skipped, obsolete, or cancelled, capturing resolution metadata, moving the record into a per-outcome archive folder, and updating the open and per-outcome INDEX.md files. Use this skill any time a user or agent declares an action is finished, no longer needed, or being abandoned — including phrasings like "mark this done", "we already did X", "skip the follow-up on Y", "Z is obsolete now", "close action <ID>", or any slash-command of the shape /record-action-taken, /resolve-action, /action-done. Also use it programmatically when a session-end agent (e.g. chat-session-scribe) detects that an open action has been completed during the session. Always use this skill — never edit the action file by hand — because the resolution must be recorded with timestamp, actor, outcome, and history, and because the file must move atomically between folders alongside two INDEX.md updates. The skill requires a caller-supplied MEMORY.md path; without it the skill

Records a decision that has just been made — architecture, product, engineering, legal, operational, or any other domain — as a closed decision file, updates the relevant index, and creates a paired action plan for tracking implementation. Captures whether the decision is `final` (no scheduled re-evaluation) or `temporary` (revisit at a known trigger such as a phase, date, threshold, or external event), and routes the file into the matching subfolder when the workspace declares one. Also handles the "confirm" path that promotes a temporary decision to final when its trigger has fired and the answer still holds. Triggers whenever the user confirms, approves, finalises, closes, resolves, or signs off on an open question, including phrases like "we've decided X", "let's go with Y", "close item A.5", "record this", "lock it in", "supersede the previous decision", "confirm A.4 as final", or "promote the temporary decision to final". Reads workspace-specific folder paths, ID series, subfolder routing, decision-type

Creates and maintains restart guides — session-handoff documents that capture the in-flight state of multi-session tasks so a future agent can resume in ~5 minutes instead of re-discovering context over 20–60 minutes. At session start, discovers active guides under `memories/restart-guides/active/` and offers a chooser. At session end, writes a new guide or updates the loaded one — capturing files modified, decisions opened/closed, action items in flight, traps, exact next-session steps — and maintains `_index.md` and the lifecycle subfolders. Use this skill whenever the user says "continue where we left off", "resume the previous session", "check if there's a restart guide", "prepare a restart guide", "write a handoff doc", "create a session handoff", invokes `/prepare-restart-guide`, or whenever the agent is starting or ending a session and a multi-session task is in flight — even if the user doesn't use those exact words.

Classifies and persists memory entries to the correct Tier 2 topical store and conditionally updates the Tier 1 MEMORY.md index via the highest-priority available MCP backend. Covers decisions, corrections, preferences, lessons, people, projects, activities, environment facts, analysis, patterns, and daily session logs. Activate proactively — without waiting to be asked — when the user says "remember this", "save this decision", or "note this preference"; or when a decision, correction, new person, or session end occurs. Companion to compact-agent-memories.

Converts plain text (.txt) files into well-structured Markdown (.md) files that strictly comply with the DavidAnson/vscode-markdownlint ruleset. Use this skill whenever the user wants to convert, transform, or reformat a text file into markdown — even if they say things like "turn this into markdown", "make this a markdown doc", "format my notes as markdown", "clean up this text file", or "convert my .txt to .md". Also trigger when the user uploads or references a .txt file and asks for any kind of document formatting, structuring, or reorganization. If a text file needs to be made readable, structured, or publishable as markdown, use this skill.

Simulates an ATS (Applicant Tracking System) compatibility score by comparing a resume against a job description. Produces a scored report covering keyword match, experience tenure, education, formatting, and prioritized improvement recommendations. Use when a user wants to know how well their resume matches a job posting, check their ATS score, optimize their resume for a specific role, or understand what keywords they are missing.

Analyzes a resume against a job description and produces a crystal-clear, action-oriented fix guide with Before/After/Rationale for each recommended ATS improvement. Use when the user wants to fix ATS issues in their resume, improve their ATS score, get specific resume edits to target a job posting, or implement recommendations from an ATS score report.

Analyzes a person's resume (PDF) and produces a structured assessment covering sentiment, skills and experience inventory, strengths, weaknesses, notable achievements, recruiter questions, and recruiter concerns. Use when the user wants to evaluate, review, score, or get feedback on a resume or CV, or when the user asks what a recruiter might think about someone's profile.

Use this skill whenever the user wants to read or extract content from a PDF file. This includes extracting text, tables, metadata, or images from an existing PDF, and performing OCR on scanned PDFs to make their content readable. Can be invoked as a sub-skill by any skill that needs to process PDF documents. Do NOT use this skill for creating, modifying, merging, splitting, rotating, watermarking, encrypting, or filling PDF files.

Assess code quality and provide structured recommendations for bugs, simplification, security, and performance. Use when the user asks to review code, clean code, assess code quality, or get recommendations for a code block in any programming language.

Adds clear, concise inline comments to code blocks to improve readability. Use when the user asks to add comments, comment code, document code, or make code more readable in any programming language.

Determines why a block of code was written and what the author was trying to achieve. Use when the user asks what code is for, why it was written, the purpose of a code block, or the intent behind code in any programming language.

Produces a markdown document that explains in plain English and in minute detail what a block of code does. Use when the user asks to explain code, document what code does, walk through code step by step, or understand what each part does in any programming language.

Identifies the programming language of a given code block and returns it as a JSON object. Use when the user asks what language code is written in, to detect or identify the programming language of a code snippet, or when another skill needs the language determined first.