srulys-agent-packs

Grill-me interrogation discipline: convert context gaps into a relentless, adversarial, gap-closure-driven question set. Defines P0/P1/P2 tagging, gap-closure termination (no tight cap; large structural safety bound), the MC-vs-freeform rubric (2-6 choices + a 'Not sure / decide later' deferral, no 'Other'), do-not-invent-answers, and the bounded re-grill loop ending in [TBD]+OQ-NN. Triggers on: interview the user, grill me, structured PRD interview, gap questions, P0 questions.

Critic scoring rubric for drafted specs. Defines dimensions D1–D4 (both modes) and D5–D8 (update mode only), per-dimension scoring guidance, weighted aggregation, verdict thresholds (pass/revise/block). Replaces literal-diff-based rubric. Triggers on: score the spec, critic rubric, D1 mandatory coverage, D2 gated appropriateness, verdict thresholds.

Adaptive PRD section catalogue and complexity heuristic. Declares each section as mandatory or complexity-gated:<axis>, with industry-neutral names. Single source of truth for which sections to include and how to name them. Triggers on: PRD section list, PRD template, complexity heuristic, mandatory vs gated sections, adaptive sectioning.

Distilled spec-driven-development and PRD best practices: PR/FAQ framing (Working Backwards), structure-before-prose, outcome-over-output, P0/P1/P2 prioritisation, testable acceptance criteria, open-questions discipline. Triggers on: PRD best practices, spec-driven development, working backwards, PR/FAQ, outcome-over-output, testable acceptance criteria.

Entry workflow for creating evalpilot evals for Copilot agents and skills. Bootstraps evals/, discovers targets, scaffolds Markdown *.eval.md or TypeScript *.eval.ts specs, and hands off to eval-runner. Trigger keywords: create evals, author evals, test my agent, test my skill, evalpilot, eval.md, eval.ts, rubric, metric, regression.

Templates and patterns for creating GitHub Copilot CLI artifacts. Use when generating custom agents or skills. Keywords: agent.md, SKILL.md, implementation.

Multi-agent system design patterns and guidance. Use when designing agent architectures, planning agent boundaries, determining communication patterns, or evaluating single vs multi-agent approaches. Keywords: architecture, topology, orchestration, agent design.

Use evalpilot numeric metrics over time: JSONL history, baseline strategies, regression gates, evalpilot metrics reports, HTML trend charts, and CI checks. Trigger keywords: eval metrics, metric regression, history.jsonl, baseline, rolling mean, tolerance, trend, evalpilot metrics.

Run evalpilot evals and triage failures for Copilot agents and skills. Teaches evalpilot run/show/lint, repo wrappers, tag filters, the modeled JSON result, terminal/HTML renders, SUT logs, exit codes, and environment knobs. Trigger keywords: run evals, evalpilot run, failing eval, eval report, triage eval, agent log.

Multi-layer codebase discovery heuristics for context packs: how to find ALL related paths across code, data, tests, docs, config, and dependencies, breadth-first, from a feature seed. Keywords: discovery, find related code, layers, code map, search heuristics, context pack.

Schema for codebase context packs: the 5 mandated content areas (with the legacy 12-section map), the confidence-scoring rubric, generated SKILL.md frontmatter, and the context-pack.json provenance schema. Keywords: context pack, content areas, confidence score, context-pack.json, schema.

Progressive-disclosure + install logic for generated context packs: the SKILL.md token-split threshold and measurement, the uniform index schema, and the write+copy-back install scripts. Keywords: progressive disclosure, split threshold, SKILL.md index, references, install script, copy-back, round-trip.

Comprehensive executive writing craft for product briefs. Decision-maker framing, persuasive structure, so what? test, tone rules, filler phrase elimination, good/bad examples, anti-repetition rules, and markdown quality. Trigger keywords: executive summary, concise writing, plain language, active voice, so what, decision-maker, persuasive.

Framework for decision-grade product briefs with brief type axis (decision-brief vs scope-brief), canonical section order, agency-over-formatting rules, section distinctness contract, right-length protocol, standalone document policy, and markdown lint compliance. Trigger keywords: product brief, decision memo, scope brief, brief type, section order, decision ask, closing section, recommendation, next steps, call to action, summary, FAQ, evidence log, brevity, standalone.

Psychological principles for making product briefs championing-ready. Covers the cascade principle, cognitive burden minimization, incentive alignment, psychological risk reduction, and championing language. Trigger keywords: stakeholder championing, psychological risk, incentive alignment, cascade principle, championing test, business outcome framing.

Maintain and evolve a Product Management knowledge base / living wiki from already-extracted text. Owns the 10-step knowledge-evolution cycle and a durable STM checkpoint so a mid-cycle context compaction loses nothing; routes to consolidation, organization, and indexing specialist skills. Triggers on: knowledge base, knowledge brain, institutional memory, product wiki, consolidate knowledge, update the knowledge base, record decision, product knowledge, evolve knowledge, living wiki, organizational memory.

Maintain discovery indexes for a Product Management knowledge base, apply refactoring heuristics at size thresholds (split oversized pages, merge duplicates, recategorize, rebuild indexes), and generate dynamic specialized index skills as routing layers when an area grows large. Specialist skill of the Product Knowledge Brain. Keywords: discovery index, knowledge index, rebuild indexes, refactor knowledge base, size threshold, dynamic index skill, routing layer, knowledge retrieval.

Own the on-disk product-centric layout of a Product Management knowledge base, the living-page templates per knowledge type, typed relationships plus backlinks, and the provenance / evidence descriptor schema. Specialist skill of the Product Knowledge Brain. Keywords: knowledge base layout, living page, page template, relationships, backlinks, wiki links, provenance, evidence descriptor, product wiki structure.

Classify already-extracted product information into the six knowledge types, determine which existing knowledge-base pages and product areas it affects, decide update / merge / correct / expand versus create (prefer consolidation over proliferation), detect contradictions against current understanding, and remove duplication. Specialist skill of the Product Knowledge Brain. Keywords: classify knowledge, consolidate, deduplicate, merge pages, update over create, contradiction detection, knowledge types.

Author Marp/Marpit markdown decks, render them to PNG/PDF for visual QA, and convert Marp → PPTX — with a self-probing, verify-or-block toolchain policy. Use when output_mode is marp or both. Keywords: marp, marpit, markdown slides, marp-cli, marp to pptx, marp theme, output_mode.

Technical foundation for programmatic PowerPoint generation using python-pptx. Covers API patterns, template mode vs default mode, slide layout strategy, styling, charts, tables, and error handling. Keywords: python-pptx, PowerPoint, pptx, slide, layout, chart, table, shape, font, generate.

Programmatic python-pptx structural assertions for deck QA. Runs fifteen checks: text-frame overflow, contrast violations (luminance ratio), alignment-grid drift, repeated-layout-hash, title-underline spam, dark/light run length, missing speaker notes, duplicate titles, body-word density, font render-presence (portability), safe-area breathing-room, plus three archetype-spec checks (waterfall zero-baseline algebra, decision_options column-width sum, risk_heatmap WCAG-AA contrast). Single entry point: check_pptx.py emits a merged JSON report consumed by @deck-critic. Keywords: structural QA, contrast, overflow, alignment grid, layout hash, archetype asserts, font portability, safe area, python-pptx assertions.

Headless render pipeline (.pptx → .pdf → per-slide .png via LibreOffice + pypdfium2) plus per-slide visual rubric (incl. the aesthetic_craft axis) for multimodal inspection. Loaded by @deck-critic to actually look at generated decks. Cross-platform with fallbacks; permissive-only engines. Keywords: render, screenshot, headless, soffice, libreoffice, pypdfium2, pdftoppm, visual QA, multimodal, rubric, aesthetic craft.

Positive visual-design rules for slides — what good looks like. Slide-type taxonomy, one-message-per-slide, visual hierarchy, layout vocabulary, sequencing patterns, color/type tokens, design quality checklist. Antipattern catalog lives in slide-critique. Keywords: slides, presentation, deck, visual hierarchy, slide types, layout, design, PowerPoint.

Code-rendered static visuals for PowerPoint decks. Three deterministic JSON-spec → PNG/SVG pipelines: matplotlib charts (bar_with_callouts, donut, sparkline, dual_bar_diff, progress_strip), PIL composites (gradient_pattern, oversized_glyph_bg), and Graphviz diagrams (flow_diagram, system_diagram). Loaded by @deck-builder; pre-renders slides[].visual_assets[] before deck assembly. Keywords: matplotlib, PIL, Pillow, Graphviz, chart, diagram, gradient, render, visual_assets, deterministic.

Curated digest of widely-used deck-design canon (Pyramid Principle, Presentation Zen, Duarte, BCG action titles, Beautiful.ai/Gamma, marp/slidev) plus ten fully-specified design systems (palette + tint ladders + type scale + grid + slide-type defaults) usable directly as deck-spec config. Loaded by @deck-builder for system selection. Keywords: design system, palette, typography, grid, theme, executive, technical, customer, investor, editorial, boardroom, ink-editorial, quiet-luxury, signal-dark, warm-editorial.

Methods for evidence quality, decision-relevance filtering, compact table formatting, source traceability without links, contradiction handling, and explicit assumption/open-question labeling. Trigger keywords: evidence quality, source traceability, contradiction resolution, confidence, assumptions, decision-relevance.

Detection algorithm for MCP servers and CLIs available in the user's environment. Defines sources (user prompt, copilot-instructions, harness tool listing), use rules, and graceful degradation. Defines how tool availability is reported through the research-runner's mcp-results fence and handoff. Triggers on: detect MCPs, MCP discovery, CLI discovery, available tools, graceful degradation.

Step 4 of the EARS PRD workflow and the quality source-of-truth. Defines the 6 EARS shall-statement patterns, nested testable acceptance criteria, the lightweight mandatory PRD section catalogue, and a pre-present self-check. Triggers on: format the PRD, EARS shall statement, acceptance criteria, PRD sections, spec quality check, requirements syntax.

Master workflow for producing a high-quality EARS-style PRD fast, as a skills-only plugin with no custom agent. Owns the 4-step flow (gather context, grill, propose outline, format) and the outline-approval loop. Triggers on: write an EARS PRD, EARS spec, draft a requirements doc, spec authoring, PRD, requirements syntax, product requirements document.

Step 2 of the EARS PRD workflow. A grill-me-style interrogation that closes every requirement gap before drafting, with NO upper cap on questions (completeness over speed). Poses questions via the ask-question ask_user tool, multiple-choice for enumerable answers and freeform for open-ended ones. Triggers on: grill me, interrogate, ask the gap questions, interview for the PRD, what is missing.

Step 1 of the EARS PRD workflow. Detects available tools (file-system search, web search, MCP servers), builds a short in-conversation context digest, and degrades gracefully when tools are missing. Triggers on: gather context, research the spec, what MCP web or file tools are available, graceful degradation, build PRD context.

Update-mode rules for evolving an existing spec: semver-for-specs (MAJOR/MINOR/PATCH), RFC-style Updates:/Obsoletes: headers, ADR-style deprecation pattern, Keep-a-Changelog categories, naming-evolution and alias conventions, living-document preamble. Triggers on: update spec, evolve spec, version bump, changelog, deprecate requirement, supersede.

Draft vs. published lifecycle for specs: mode model, branch-aware mode inference, publish transition, post-publish numbering stability, re-draft window, cross-reference integrity, status surfacing, malformed-front-matter handling, and the publish-time changelog rule. Single source of truth for the V1–V18 normative rules. Triggers on: draft mode, published, version bump, publish, re-draft, renumber FR, branch detection, status field, version field, pre-merge reminder, CHANGELOG.

Storytelling expertise for transforming graph-like product knowledge into compelling linear narratives. Covers storytelling frameworks (SCR, Hero's Journey, Problem-Solution), audience-framework selection, narrative arc design, and the So What test. Keywords: storytelling, narrative, story arc, framework, audience, SCR, hero journey, graph-to-linear.

Negative-space catalog and checklist execution for slide-deck QA. Catalogs AI-generated slide antipatterns, defines the critique output format, and orchestrates execution through the @deck-critic visual-QA pipeline (render + structural-asserts + multimodal rubric). Keywords: critique, audit, review, design quality, visual QA, feedback loop, antipatterns.

Detection algorithm for MCP servers and CLIs available in the user's environment. Defines sources (user prompt, copilot-instructions, harness tool listing), use rules, and graceful degradation. Provides the canonical discovery.json schema. Triggers on: detect MCPs, MCP discovery, CLI discovery, available tools, graceful degradation.

Comprehensive guide for building agents, skills, and prompts for GitHub Copilot CLI. Use this skill when creating custom agents (.agent.md files), skills (SKILL.md files), or custom instructions for Copilot CLI. Covers tool reference, file structure, best practices, and prompt engineering patterns.

Business logic detection patterns for identifying authorization, events, domain models, and API mappings across common frameworks. Use when analyzing code to extract business-relevant knowledge. Keywords: business logic, authorization, RBAC, events, domain model, API mapping.

Knowledge base template structure, section rules, dual-layer formatting, and quality lint rules for code intelligence output. Use when synthesizing findings into KB documents or validating KB output. Keywords: KB template, knowledge base, dual-layer, formatting, section structure.