Skip to content

Requirements

These requirements cover the RedTeam skills package, CLI installer, provider build copies, project .redteam/ scaffold, and developer documentation sync. They translate validated experience artifacts in experience/ into a delivery contract engineering can test and publish.

ID Requirement Derived from Acceptance behavior
FR-1 The system shall route /redteam <command> through skill/SKILL.md to the matching skill/reference/<command>.md flow. experience/chat-only-review.md Given an installed skill, when the user invokes a valid command, then the model follows that command’s reference flow.
FR-2 The system shall produce structured adversarial review output for challenge-style commands without requiring CONTEXT.md. experience/chat-only-review.md, PRODUCT goals Given no CONTEXT.md, when the user runs /redteam challenge <target>, then output includes assumptions, risks, a verdict, and follow-ups.
FR-3 The system shall write decision context via /redteam init to CONTEXT.md and scaffold .redteam/. experience/project-anchored-review.md Given a project directory, when the user runs /redteam init, then CONTEXT.md and .redteam/ scaffold exist.
FR-4 The system shall load decision context via scripts/context.mjs using resolution order: root CONTEXT.md → .redteam/.agents/context/docs/$REDTEAM_CONTEXT_DIR. experience/project-anchored-review.md Given CONTEXT.md in a supported location, when context.mjs runs, then context content is available to the command flow.
FR-5 The system shall persist project-anchored review output to .redteam/reviews/<slug>-<command>.md. experience/project-anchored-review.md Given project-anchored mode, when a review command completes, then a markdown artifact exists under .redteam/reviews/.
FR-6 The CLI installer shall copy the skill into requested provider directories and create .redteam/config.json and .redteam/CONTEXT.template.md when missing. experience/install-and-run.md Given a clean directory, when install runs, then expected skill paths and .redteam/ scaffold files exist.
FR-7 The build script shall copy skill/ to all configured provider destinations without manual editing. experience/install-and-run.md Given source skill, when npm run build runs, then provider skill directories match source content.
FR-8 The developer docs site shall stage every docs/**/*.md file and resolve internal links for GitHub Pages under /RedTeam/developers/. engineering/TESTING.md, README published site Given docs/ content, when npm run docs:site runs, then the build succeeds and local links resolve under the base path.
FR-9 The system shall implement 29 handbook-aligned /redteam commands documented in the user guide. PRODUCT goals Given the command table in skill/SKILL.md and the user guide command list, when each command is invoked, then a corresponding reference flow exists in skill/reference/.
FR-10 The system shall support pin / unpin management commands so $<command> invokes /redteam <command> directly. skill/SKILL.md pin flow Given an installed skill, when the user runs /redteam pin <command>, then pin.mjs creates the shortcut in the harness directories and reports the result.
FR-11 The system shall persist facilitation session notes to .redteam/sessions/<date>-<slug>.md. experience/project-anchored-review.md, skill/SKILL.md output conventions Given project-anchored mode, when a facilitation command (groupthink, ideate, converge) completes, then a dated markdown artifact exists under .redteam/sessions/.
FR-12 Every handbook chapter shall teach through an HBS-style case-study narrative that places concepts and TTPs inside a consequential decision; the chapters shall build cumulatively and the final chapter shall culminate in the complete “spell book.” PRODUCT handbook goal, DESIGN content structure Given any chapter, when a reader enters it, then the chapter presents stakes, incomplete information, competing perspectives, and a decision before or alongside instruction; given the final chapter, then the accumulated concept and TTP repertoire is available as a browsable working reference.
FR-13 The handbook shall support both sequential reading and direct concept/TTP lookup. PRODUCT handbook goal, DESIGN two-ways-in principle Given the handbook landing page, when a reader arrives, then they are invited either to begin the cumulative case journey or to navigate directly to concepts and TTPs.
ID Quality attribute Target / constraint Why it matters
NFR-1 Setup friction Install to first command in under 5 minutes Adoption goal in experience/install-and-run.md
NFR-2 Runtime platform Node >=22.12.0 for CLI and skill scripts Documented engine constraint
NFR-3 Security Skill scripts make no network calls; no exploit or penetration-testing content Civilian product boundary
NFR-4 Portability Skill works in Claude, Cursor, Codex, ChatGPT, Copilot, and Gemini harnesses Multi-harness distribution goal
NFR-5 Attribution Handbook techniques summarized, not reproduced verbatim Legal and lineage policy in strategy/source-lineage.md
NFR-6 Offline capability Chat-only commands work without network after skill install Pure-chat experience
NFR-7 Method fidelity Evaluative command output names the technique applied and separates evidence, interpretation, assumptions, and uncertainty PRODUCT quality stance: methods, not generic criticism; evidence before confidence
NFR-8 Accountability High-stakes command output states important uncertainties and leaves the decision with its human owner PRODUCT quality stance: human judgment remains accountable
Requirement Given When Then
FR-2 No CONTEXT.md /redteam challenge <plan> Structured review in chat
FR-5 CONTEXT.md exists /redteam premortem launch .redteam/reviews/launch-premortem.md saved
FR-6 Clean temp project redteam install --providers=agents Skill + .redteam/ scaffold on disk
FR-8 docs/ tree complete npm run docs:site Build passes; /RedTeam/-prefixed links valid
FR-10 Skill installed /redteam pin premortem $premortem shortcut exists in harness dirs
FR-11 .redteam/ exists /redteam groupthink <session> .redteam/sessions/<date>-<slug>.md saved
FR-12 Any handbook chapter Reader follows the chapter A decision case frames the concepts and TTPs; the final chapter exposes the full spell book
FR-13 Reader opens the handbook Reader chooses an entry path Sequential chapter and direct concept/TTP routes are both explicit and usable
NFR-7 Any evaluative command Output reviewed Technique named; evidence, interpretation, assumptions, and uncertainty distinguishable
  • Constraint: Skills-only — no hosted API, database, or model calls from repository code
  • Constraint: Apache-2.0 license; NOTICE and lineage docs must remain accurate
  • Assumption: Users bring a capable LLM harness that follows skill instructions
  • Assumption: Quality of review output depends on model adherence to reference flows (no deterministic validator)
  • Node.js 22.12+ — CLI, build, context loading, docs site toolchain
  • GitHubnpx github:CurateLabs/RedTeam distribution and Pages hosting
  • Astro / Starlight — published handbook, guide, and developer docs
  • AI harness platforms — Cursor, Claude, Codex, ChatGPT, Copilot, Gemini for skill execution
  • Which commands should gain automated output-shape evals beyond install/docs smoke tests? — maintainers
  • How are NFR-7 / NFR-8 judged in practice — sampled-session rubric, expert review, or both? (See PRODUCT “Evidence of success”.) — maintainers