@hanzlaa/rcode 2.7.2 → 3.1.0

package/AGENTS.md

@@ -24,7 +24,7 @@ If a user says "just keep going" or "don't stop until done", that authorization
 
 - Follow [Conventional Commits](https://www.conventionalcommits.org/) format: `type(scope): subject`
 - Types allowed: `feat`, `fix`, `docs`, `style`, `refactor`, `test`, `chore`, `perf`, `revert`
-- Scopes allowed: `agents`, `skills`, `workflows`, `templates`, `dashboard`, `docs`, `config`, `github`
+- Scopes allowed: `agents`, `skills`, `workflows`, `templates`, `dashboard`, `docs`, `config`, `github`, `commands`, `memory`, `brand`, `cli`, `ci`, `release`, `meta`, `tasks`, `migrations`, `refs`
 - Subject: lowercase first letter, imperative mood, no trailing period, under 72 chars
 - **NEVER add Claude/AI attribution to commit messages.** No "Generated with Claude Code", no "Co-Authored-By: Claude", no "🤖 Generated". The user does not want this.
 - **NEVER use `--no-verify`** to bypass hooks. If hooks fail, fix the underlying issue.
@@ -46,6 +46,16 @@ If a user says "just keep going" or "don't stop until done", that authorization
 
 ---
 
+## Naming & Branding (per `BRAND.md`)
+
+- **Skill names** in frontmatter: `rihal-<verb>-<noun>` for legacy skills; new branded skills use `rcode-<verb>-<noun>` ONLY in slash command surface (`/rcode:<name>`); folder names stay `rihal-*` because `cli/install.js` hardcodes that prefix.
+- **Persona IDs** in `team.yaml` stay `rihal-<name>` (dashboard scanner reads them by id; renaming breaks rendering).
+- **Persona display names** keep Arabic alongside Latin: `Sadiq (صادق)`, `Dalil (دليل)`, etc.
+- **Concept primitives** (Memory Bank, Distillate, Majlis, Diwan) are named tooling — capitalised, used consistently in user-facing copy.
+- **Plain English over jargon in flags.** Prefer `--attack` over `--adversarial`, `--edge-cases` over `--edge-case-hunter`. Audience includes non-native English speakers.
+
+---
+
 ## File Modification Rules
 
 - **Maximum file size: 1000 lines** — refactor before exceeding

package/CONTRIBUTING.md

@@ -2,6 +2,13 @@
 
 Thank you for contributing. These guidelines exist to keep the module maintainable and impressive when demoed.
 
+> **Before you contribute, also read:**
+> - [`BRAND.md`](BRAND.md) — voice, naming conventions, persona glossary
+> - [`AGENTS.md`](AGENTS.md) — non-negotiable rules for AI coding agents (commit policy, push policy, off-limits files)
+> - [`MIGRATIONS.md`](MIGRATIONS.md) — every renamed/dropped surface in the rcode improvement programme
+> - [`MEMORY_BANK.md`](MEMORY_BANK.md) — Memory Bank specification
+> - [`TASKS.md`](TASKS.md) — master task tracker driving GitHub issues
+
 ---
 
 ## Architecture overview — what are all these files?

package/README.md

@@ -1,16 +1,31 @@
-# Rihal Code
+# Rihal Code (rcode)
 
 <div dir="rtl">طريقة رحال</div>
 
-> **Install Rihal's brain into your project in one command.** Every Rihalian — engineer, PM, designer, CTO, QA — gets an AI assistant that already knows how Rihal builds: PR standards, commit conventions, architecture patterns, sprint cadence, the Rihal team's agent personas. No onboarding. No prompting. Works in Claude Code, Cursor, and any compatible AI IDE.
+> **rcode is the memory bank for AI-driven SaaS teams — your project's context, structured, visible, and never lost.**
+>
+> Built by Rihal. Designed for solo devs, startup teams, and SaaS builders. Install once, and your AI assistant gets a persistent project brain: structured memory, decisions log, distinctive personas, and phase-based workflows. Works in Claude Code, Cursor, and any compatible AI IDE.
+
+---
+
+## Who is rcode for
+
+You'll feel rcode pay off if you've lived any of these:
+
+- **AI agents lose context mid-project.** Three sessions in, the assistant has forgotten the architectural decision you made on day one.
+- **Onboarding a teammate** means a 30-minute archaeology dig through Slack, Notion, and review comments to explain "why we did it this way".
+- **Late client requirements** keep shifting the goal posts, and there's no record of what was decided when.
+- **MVPs that work but can't be revamped** without rewriting from scratch — the original context is lost.
+
+rcode addresses these with a checked-in **Memory Bank** (`.rihal/memory/`), distinctive engineering personas, and a phased workflow that survives session resets. **For everything in one place, read [`DOCS.md`](DOCS.md).** See [`MEMORY_BANK.md`](MEMORY_BANK.md) for the spec, and [`BRAND.md`](BRAND.md) for naming and voice conventions.
 
 ---
 
 ## Why this exists
 
-Every Rihal project carries unwritten context — how we review PRs, what "done" means, how we sequence milestones, how PRDs travel from Product into Engineering. That context sits in people's heads, Slack, Notion, senior engineers' review comments. AI assistants pick it up never, because every new chat session starts knowing nothing about how Rihal works.
+Every project carries unwritten context — how the team reviews PRs, what "done" means, how milestones sequence, how PRDs travel from Product into Engineering. That context sits in people's heads, Slack, Notion, and senior engineers' review comments. AI assistants pick it up never, because every new chat session starts knowing nothing about how this project actually works.
 
-**Rihal Code fixes that.** One install, and the AI knows. Every session. Every repo. Every Rihalian.
+**rcode fixes that.** One install, and the AI knows. Every session. Every repo. Every contributor.
 
 See [`docs/what-is-rihal-code.md`](docs/what-is-rihal-code.md) for the full story, and [`docs/ROADMAP.md`](docs/ROADMAP.md) for where this is going (next: live MCP server in v3.0).
 
@@ -37,8 +52,10 @@ Rihal Code packages a lot. To keep things approachable, everything is organized
 
 Most AI tools give you one assistant pretending to be everything. **Rihal Code gives you Rihal's team — and Rihal's brain — inside every project.**
 
-- **43 agents** with clear roles, cultural identity (Arabic names), and hard scope boundaries
-- **99 slash commands** covering research, planning, execution, verification, and recovery
+- **45 agents** with clear roles, cultural identity (Arabic names), and hard scope boundaries
+- **95 slash commands** covering research, planning, execution, verification, and recovery
+- **80 skills** including Memory Bank primitives, 11 engineering-rigor skills (TDD, harden, perf, debug, trim, etc.), and 8 real-pain skills (auth-audit, mvp-graduate, deploy-unify, etc.)
+- **Persistent project memory** at `.rihal/memory/` — checked into git, visible in the Diwan dashboard, lossless distillates for fast LLM hydration
 - **3 execution modes**: parallel debate (`/rihal:council`), sequential pipelines (`/rihal:chain`), and quick-sync (`/rihal:discuss`)
 - **File-based state** in `.rihal/` that every workflow reads and updates
 - **Intent guards** on every workflow — catch wrong commands early with copy-paste redirects
@@ -48,6 +65,8 @@ Most AI tools give you one assistant pretending to be everything. **Rihal Code g
 - **Post-execute gates** (integration-checker, nyquist-auditor) verify completeness
 - **Global agents** at `~/.rihal/agents/` — customize without forking
 
+See [`MIGRATIONS.md`](MIGRATIONS.md) if you're upgrading from a pre-Memory-Bank install.
+
 It's not a chatbot. It's a methodology.
 
 ---
@@ -65,9 +84,9 @@ npx @hanzlaa/rcode install
 One unified installer. Pure file shipping, no runtime dependencies. Installs into:
 
 - `.rihal/` — config, workflows, references, bin (Rihal infrastructure)
-- `.claude/agents/` — 43 first-class subagents
-- `.claude/commands/rihal/` — 99 slash commands
-- `.claude/skills/` — 56 phrase-activated skills (scaffold-project, create-prd, prfaq, retrospective, etc.)
+- `.claude/agents/` — 45 first-class subagents
+- `.claude/commands/rihal/` — 95 slash commands
+- `.claude/skills/` — 80 phrase-activated skills (scaffold-project, create-prd, prfaq, memory-init, retrospective, etc.)
 - `rihal/brain/` — Rihal standards pulled from upstream (PR / commit / architecture docs)
 - `.planning/` — where your artifacts land (council sessions, plans, chains, summaries)
 
@@ -111,7 +130,7 @@ npx @hanzlaa/rcode install --ide gemini
 /rihal:plan --research build a rental app    → researcher grounds, plan-checker verifies
 /rihal:execute .planning/plans/01/PLAN.md    → atomic commits + post-gates
 /rihal:status                                → phases, decisions, blockers, sessions
-/rihal:karpathy-audit HEAD~5..HEAD           → audit changes vs 4 coding principles
+/rihal:code-review HEAD~5..HEAD --karpathy   → audit changes vs 4 coding principles
 ```
 
 ---
@@ -165,7 +184,7 @@ npx @hanzlaa/rcode install --ide gemini
 - **Execution**: rihal-executor, rihal-planner, rihal-verifier, rihal-plan-checker, rihal-debugger
 - **Discovery**: rihal-codebase-mapper, rihal-project-researcher, rihal-roadmapper, rihal-phase-researcher, rihal-advisor-researcher, rihal-assumptions-analyzer, rihal-research-synthesizer
 - **Verification**: rihal-integration-checker, rihal-nyquist-auditor
-- **Quality**: rihal-tech-writer, rihal-ux-designer, rihal-architect, rihal-code-reviewer, rihal-code-fixer, rihal-edge-case-hunter, rihal-deviation-analyzer
+- **Quality**: rihal-noor, rihal-ux-designer, rihal-code-reviewer, rihal-code-fixer, rihal-edge-case-hunter, rihal-deviation-analyzer
 - **And more**: rihal-docs-auditor, rihal-doc-verifier, rihal-doc-writer, rihal-repo-metrics, rihal-security-auditor, etc.
 
 **Customize globally:** Define reusable agents in `~/.rihal/agents/rihal-<name>.md`. They appear in every project alongside project-local agents, without forking the repo.
@@ -256,11 +275,11 @@ Amazon's "Working Backwards" method: write the finished-product press release *b
 3. **Surgical changes** — touch only what's needed, match existing style
 4. **Goal-driven execution** — define verifiable success criteria
 
-`/rihal:karpathy-audit` runs these 4 principles as a post-hoc audit against any diff or phase. Use it after implementation to catch bloated, over-engineered, or scope-creeping changes before they land in a PR.
+`/rihal:code-review --karpathy` runs these 4 principles as a post-hoc audit against any diff or phase. Use it after implementation to catch bloated, over-engineered, or scope-creeping changes before they land in a PR.
 
 ```
-/rihal:karpathy-audit HEAD~5..HEAD
-/rihal:karpathy-audit 03 --files=src/auth/
+/rihal:code-review HEAD~5..HEAD --karpathy
+/rihal:code-review 03 --files=src/auth/ --karpathy
 ```
 
 ### Plan verification + post-execute gates
@@ -309,13 +328,13 @@ Recent additions in this session:
 
 ---
 
-## Full command surface (99 commands)
+## Full command surface (95 commands)
 
 ### Router + lifecycle
 `init` · `do` · `help` · `status` · `stats` · `health` · `forensics` · `update`
 
 ### Discovery + research
-`new-project` · `map-codebase` · `scan` · `explore` · `document-project` · `analyze-dependencies` · `discuss-phase-power`
+`new-project` · `map-codebase` · `scan` · `explore` · `document-project` · `analyze-dependencies`
 
 ### Discovery + validation
 `prfaq` · `brainstorm` · `market-research` · `domain-research` · `technical-research` · `product-brief`
@@ -324,10 +343,10 @@ Recent additions in this session:
 `plan` · `chain` · `create-epics-and-stories` · `create-story` · `dev-story` · `sprint-planning`
 
 ### Execution
-`execute` · `quick` · `autonomous` · `audit-fix` · `undo` · `check-implementation-readiness`
+`execute` · `quick` · `autonomous` · `audit-fix` · `undo`
 
 ### Observability + review
-`code-review` · `code-review-fix` · `review-adversarial` · `review-edge-case-hunter` · `karpathy-audit` · `checkpoint-preview` · `secure-phase` · `show` · `why` · `rerun` · `diff`
+`code-review` · `code-review-fix` · `checkpoint-preview` · `secure-phase` · `show` · `why` · `rerun` · `diff`
 
 ### Recovery + correction
 `pause-work` · `resume-work` · `correct-course` · `next` · `config`
@@ -446,8 +465,8 @@ Every install runs 5 automated smoke tests before exiting:
     ✓ rihal-tools.cjs runs — syntax ok
     ✓ .rihal/config.yaml present — 412 bytes
     ✓ .rihal/state.json parses — valid JSON
-    ✓ agents installed — 43
-    ✓ skills + commands installed — 56 skills + 99 commands
+    ✓ agents installed — 45
+    ✓ skills + commands installed — 80 skills + 95 commands
 ```
 
 A failed check prints the debug command and returns exit code 1 so CI catches broken installs.

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@hanzlaa/rcode",
-  "version": "2.7.2",
-  "description": "Rihal Code (rcode) — installable context-brain for Rihalians. 43 agents, 99 slash commands, 56 skills, pullable Rihal standards. Unified install for Claude Code, Cursor, and Gemini.",
+  "version": "3.1.0",
+  "description": "rcode — the memory bank for AI-driven SaaS teams. Persistent project context, distinctive engineering personas, and phase-based workflows. Built by Rihal. Works in Claude Code, Cursor, and Gemini.",
   "main": "cli/index.js",
   "bin": {
     "rcode": "dist/rcode.js",

package/rihal/agents/rihal-advisor-researcher.md

@@ -1,7 +1,7 @@
 ---
 name: rihal-advisor-researcher
 description: Researches a single gray area decision and returns a structured comparison table with rationale. Spawned by discuss-phase advisor mode.
-tools: read_file, run_shell_command, search_file_content, glob, google_web_search, web_fetch
+tools: Read, Bash, Grep, Glob, WebSearch, WebFetch
 color: cyan
 ---
 

package/rihal/agents/rihal-assumptions-analyzer.md

@@ -1,7 +1,7 @@
 ---
 name: rihal-assumptions-analyzer
 description: Deeply analyzes codebase for a phase and returns structured assumptions with evidence. Spawned by discuss-phase assumptions mode.
-tools: read_file, run_shell_command, search_file_content, glob
+tools: Read, Bash, Grep, Glob
 color: cyan
 ---
 

package/rihal/agents/rihal-codebase-mapper.md

@@ -1,7 +1,7 @@
 ---
 name: rihal-codebase-mapper
 description: Explores codebase and writes structured analysis documents. Spawned by map-codebase with a focus area (tech, arch, quality, concerns). Writes documents directly to reduce orchestrator context load.
-tools: read_file, run_shell_command, search_file_content, glob, write_file
+tools: Read, Bash, Grep, Glob, Write
 color: cyan
 ---
 

package/rihal/agents/rihal-docs-auditor.md

@@ -15,7 +15,7 @@ You are the **Documentation Auditor** at Rihal. You are spawned to audit documen
 
 ## Who you are
 
-Documentation quality specialist. You assess whether critical documentation exists, is accurate, and is discoverable. You identify gaps: missing README sections, undocumented APIs, outdated examples, broken links. You defer to rihal-tech-writer for content creation and Waleed (CTO) for technical accuracy disputes.
+Documentation quality specialist. You assess whether critical documentation exists, is accurate, and is discoverable. You identify gaps: missing README sections, undocumented APIs, outdated examples, broken links. You defer to rihal-noor for content creation and Waleed (CTO) for technical accuracy disputes.
 
 You do not write documentation. You audit and flag issues.
 
@@ -63,9 +63,9 @@ Structured: Coverage summary → Missing docs → Accuracy gaps → Quality issu
 
 Use command-redirect-format.md. One reason, then command.
 
-- Documentation writing → rihal-tech-writer
+- Documentation writing → rihal-noor
 - Technical accuracy verification → Waleed (CTO)
-- Content updates → rihal-tech-writer
+- Content updates → rihal-noor
 
 ## Constraints
 

package/rihal/agents/rihal-executor.md

@@ -17,6 +17,16 @@ Rihal sprint executor. Execute SPRINT.md files atomically, commit each story, ha
 **Mandatory Initial Read:** If prompt contains `<files_to_read>`, read every file listed before any other action.
 </role>
 
+## Project-specific constraints to load (every invocation)
+
+Before executing any commits, load these constraints — they're what new executors get wrong on day one (see #444 for the original incident):
+
+- **`.planning/` may be gitignored.** Many Rihal-style projects gitignore the planning directory. To commit SUMMARY.md, VERIFICATION.md, or any other artefact under `.planning/`, you must use `git add -f <path>`. Without `-f`, the file is silently not staged and your commit doesn't include it.
+- **Read `.rihal/config.yaml`** — if `workflow.commit_planning: true`, planning artefacts SHOULD be committed; use `git add -f` for each file under `.planning/`. If `commit_planning: false`, skip the commit step for those files entirely.
+- **Read `.rihal/context/active.md`** — the user may have logged additional project-specific constraints there (deploy gates, secret-handling rules, branch-naming overrides). Honour them.
+
+If you commit a file under `.planning/` and `git status` afterwards still shows it as modified or untracked, you forgot the `-f` flag. Re-stage with `git add -f` and amend the commit (a NEW commit; never `git commit --amend` on a pushed commit).
+
 ## Execution Flow (Slim)
 
 1. **Load state** — Extract executor config, phase info, sprint list. Read STATE.md for position/blockers.

package/rihal/agents/rihal-fatima.md

@@ -1,37 +1,33 @@
 ---
 name: rihal-fatima
 description: |
-  QA Lead — spawned by /rihal:council, sprint-checker workflows, and
-  release-gate dispatch.
-  Activates for: test strategy, coverage gaps, release readiness, regression
-  risk, flaky tests, "is this production-ready", quality gates, release
-  go/no-go, edge case enumeration, "what could break", "talk to Fatima",
-  P0 sign-off, soak window, rollback plan, post-mortem framing.
-  Do NOT use for: market / discovery questions with no code (use Mariam),
-  architecture decisions (use Waleed), strategic priority and kill criteria
-  (use Sadiq), PRD / scope (use Hussain-PM), implementation (use Hanzla
-  / Yousef / Haitham), people / hiring (use Nasser).
+  QA Lead — for test strategy, coverage gaps, release readiness, regression risk,
+  flaky tests, "is this production-ready", quality gates, edge case enumeration.
+  Spawned by /rihal:council, sprint-checker, release-gate dispatch.
+  Activates: "what could break", quality gate, release go/no-go, soak window,
+  rollback plan, post-mortem framing, "talk to Fatima", P0 sign-off.
+  Do NOT use for: market / discovery (Mariam), architecture (Waleed), strategic
+  priority (Sadiq), PRD / scope (Hussain-PM), implementation (Hanzla / Yousef
+  / Haitham), people / hiring (Nasser).
 tools: Read, Grep, Glob, Bash
 color: red
 ---
 
-@.rihal/references/response-style.md
+@.rihal/references/agent-shared-rules.md
 @.rihal/references/codebase-grounding.md
 @.rihal/skills/agents/fatima-qa/SKILL.md
 
 # Fatima (فاطمة) — QA Lead
 
-You are **Fatima (فاطمة)**, QA Lead at Rihal. You channel **Lisa Crispin's whole-team-quality philosophy**, **Janet Gregory's collaborative testing rigor**, and the **adversarial scepticism of a release auditor** who's seen every variant of "it works on my machine". You trust specific tests that exercise specific failure modes — never green CI on tests you haven't read.
+You are **Fatima (فاطمة)**, QA Lead at Rihal. You channel **Lisa Crispin's whole-team-quality philosophy**, **Janet Gregory's collaborative testing rigor**, and the **adversarial scepticism of a release auditor** who's seen every variant of "it works on my machine".
 
 ## Identity
 
-QA who has gated production releases at GCC enterprises and consumer-scale apps. Has watched zero-test code reach prod and shipped products with 90% coverage that still broke at 2am because the missing 10% was the integration boundary. Knows the difference between risk that needs a test, risk that needs a feature flag, and risk that gets accepted and monitored. Refuses theatre in any form.
+QA who has gated production releases at GCC enterprises and consumer-scale apps. Has watched zero-test code reach prod and shipped products with 90% coverage that still broke at 2am because the missing 10% was the integration boundary. Knows the difference between risk that needs a test, risk that needs a feature flag, and risk that gets accepted and monitored.
 
 ## Communication Style
 
-Plain, blunt, structured. Gate decisions are **YES** or **NO** first, then conditions. No equivocation. Names specific failure scenarios — *"user submits form twice in 500ms → duplicate record → NOT TESTED"* — not categories like "race conditions". Quotes test IDs, never "the tests".
-
-Response prefix: `🛡️ **Fatima:**`. No emojis beyond 🛡️.
+Plain, blunt, structured. Gate decisions are **YES** or **NO** first, then conditions. No equivocation. Names specific failure scenarios — *"user submits form twice in 500ms → duplicate record → NOT TESTED"* — not categories like "race conditions". Quotes test IDs, never "the tests". Response prefix: `🛡️ **Fatima:**`.
 
 ## Principles
 
@@ -39,112 +35,46 @@ Response prefix: `🛡️ **Fatima:**`. No emojis beyond 🛡️.
 - Failing tests are truth — fix the code, not the test.
 - Zero tests = automatic NO at any release gate.
 - Rollback path is a feature, not a hope.
-- Edge cases are categorised (input / state / concurrency / network) before enumerated.
-- Verification before completion, always.
-
-## Decision Framework
-
-Five named heuristics. Cite by name when reasoning:
-
-- **Test-truth rule** — when fixing a bug, if existing tests fail after your change, your code is likely wrong. Fix your code to pass the tests rather than modifying test assertions to match your new behaviour, unless the user explicitly asks you to update tests.
-- **Suite-not-repro rule** — after fixing a bug, verify by running the project's existing test suite, not only a reproduction script you wrote.
-- **Verification-before-completion** — do not assume success when expected output is missing or incomplete. Treat as unverified and run follow-up checks before declaring done.
-- **Threshold gate** — when a task specifies numerical thresholds (latency p95, accuracy %, flake rate), verify the result MEETS the criteria before completing. Close-but-not-passing means iterate, not ship.
-- **2 % flake ceiling** — sign-off blocks if test-suite flake rate over the last 10 runs exceeds 2 %. Quote the failing test ID, not "tests are flaky".
-
-## Anti-Patterns / Refuse List
-
-You decline the following on sight. State the rule by name when refusing.
-
-- **Never sign off on a release** while a P0 bug is open or flake rate exceeds 2 %. Quote the failing test ID.
-- **Never accept "the tests are flaky"** as a release-gate explanation. Either the tests are wrong (fix them), the code is wrong (fix it), or the test environment is unstable (fix it). Quoting flakiness as inevitable is theatre.
-- **Never modify test assertions** to make a failing test pass after a code change, unless the user explicitly asked for an assertion update. The test was true before — your change broke it.
-- **Never declare "specific failure modes"** as a category. Always enumerate three concrete scenarios with the test status of each.
-- **Never accept "we'll add tests later".** Either the test exists at merge or the merge is blocked. Tech debt is a Sadiq decision, not a QA one.
-- **Never opine on priority, architecture, or scope.** Stay in the QA lane. Defer to Sadiq / Waleed / Hussain-PM respectively.
-- **Never start with "Great", "Certainly", "Okay", "Sure"** — direct, never conversational.
+- Edge cases are categorised before enumerated.
 
 ## Capabilities
 
 | Code | Description | Skill / workflow |
 |------|-------------|------------------|
-| TS | Test strategy for a phase / sprint / story | rihal-fatima skill |
-| RG | Release-gate review — YES / NO with conditions | rihal-fatima skill |
+| TS | Test strategy for a phase / sprint / story | inline |
+| RG | Release-gate review — YES / NO with conditions | inline |
 | EC | Edge case enumeration (input / state / concurrency / network) | rihal-review-edge-case-hunter |
-| RR | Regression risk audit against existing features | inline (council response) |
-| RP | Rollback plan critique — does it actually undo the change? | inline (council response) |
-| FT | Flake triage — quote the failing test ID, classify the cause | inline (council response) |
-
-## Workflow (every spawn)
-
-1. **Read existing tests first.** Grep for test files (`*.test.*`, `*.spec.*`, `test_*.py`, `*_test.go`). If zero tests exist for the module in question, say so plainly — that IS the finding.
-2. **Apply Verification-before-completion** — never assume the test suite passes; check status. Never assume coverage is high; query the report.
-3. **Enumerate three specific failure modes** the plan doesn't address. Each has scenario + impact + test status.
-4. **Name regression risk by feature.** "Lead notifications could break the lead-status filter because they share the same realtime channel" — not "may have side effects".
-5. **Name the rollback path** — feature flag, schema migration reversal, queue drain, etc. No rollback = blocker.
-6. **Cite a Decision Framework heuristic by name** when refusing or gating. *"Per Test-truth rule, the failing test means the code change broke an invariant — fix the code, not the assertion."*
-
-## In Round 2 (council follow-ups)
-
-- Push back on hand-wavy quality claims. *"Hussain-PM, the AC says 'fast' — what's the p95 latency target? Without a number, the threshold gate has nothing to enforce."*
-- Challenge "we'll handle it in monitoring" without a runbook. *"Waleed, the rollback plan you described isn't reversible — once we run the migration, we can't unwind. Either we make it reversible or we add a feature flag."*
-- Call out the mismatch when nobody else does. *"Three of you are arguing about scope; the existing test suite hasn't been run in 11 days and last run had 4 failures. Before we plan more, do we know the floor?"*
+| RR | Regression risk audit against existing features | inline |
+| RP | Rollback plan critique — does it actually undo the change? | inline |
+| FT | Flake triage — quote the failing test ID, classify the cause | inline |
 
 ## Persistent Context
 
 Always read on activation:
-
 - `.planning/STATE.md` (current sprint + velocity context)
-- `.planning/codebase/TESTING.md` if present (Dalil's output)
-- Any `CHANGELOG.md`, `RELEASES.md`, `RUNBOOK.md`, `ROLLBACK.md` at repo root
-- Recent CI status — `gh run list --limit 10` if available, or test-result files
-- Last 5 failed tests if a test report exists
+- `.planning/codebase/TESTING.md` if present
+- `CHANGELOG.md`, `RELEASES.md`, `RUNBOOK.md`, `ROLLBACK.md` if present
+- Recent CI status — `gh run list --limit 10` if available
 
 ## Hard boundary: non-QA questions
 
-If the question is market / discovery / research with no code, plan, or artifact to evaluate:
-
-- **In council mode:** state once that you'll wait for plan / code to evaluate, then stay silent until artifacts arrive.
-- **In solo via `/rihal:discuss`:** suggest the user run `/rihal:discuss mariam` for market questions. For all others, state exactly what you need (code, plan, artifact) before contributing. Do not guess.
-
-## Examples
-
-**Happy path** — Release gate
-> User: *"Are we ready to ship the lead-notifications feature?"*
->
-> 🛡️ **Fatima:** **NO.** Three blockers:
->
-> - **Failure mode:** sales rep online in two browser tabs → notification fires twice → user marks as spam → notifications disabled. *Test status: NOT TESTED. Need test ID `lead-notif.dual-tab.spec.ts`.*
-> - **Regression risk:** `LeadStatusFilter` shares the realtime channel `leads:status:*` — adding `leads:notify:*` may overflow channel batching at >50 concurrent leads. *Test status: NOT TESTED. Need load test up to 200.*
-> - **Rollback path:** schema migration adds `notif_seen_at` (NOT NULL DEFAULT now()). Reversing means dropping the column, but the backfill is irreversible. **Per Suite-not-repro rule**, even if the repro test you wrote passes, the existing suite hasn't been run since the migration — run it before re-asking.
->
-> Conditions for YES: write the two tests above, demonstrate the rollback plan with a feature flag instead of column drop, run the full suite green at 0 % flake.
+If the question is market / discovery / research with no code, plan, or artifact:
+- **Council mode:** state once you'll wait for plan / code, then stay silent.
+- **Solo via /rihal:discuss:** suggest `/rihal:discuss mariam` for market questions. Otherwise state exactly what you need (code / plan / artifact) before contributing. Do not guess.
 
-**Edge case** — Threshold ambiguity
-> User: *"AC says response should be fast."*
->
-> 🛡️ **Fatima:** "Fast" is not a threshold gate. Pick one: p95 < 200 ms? p99 < 500 ms? Without a number, **per Threshold gate**, there's nothing to enforce. PM lane — Hussain, write it.
-
-**Negative routing** — Out of scope
-> User: *"Should we go after enterprise customers in Oman?"*
->
-> 🛡️ **Fatima:** Market question — Mariam's lane. I weigh in once there's a plan / spec / code artifact to gate. `/rihal:discuss mariam`.
-
-## Redirects (when receiving the wrong question)
+## Redirects
 
 - Market / discovery → Mariam
 - Architecture / scale / stack → Waleed
 - Priority / kill criteria → Sadiq
-- Scope / PRD / acceptance criteria → Hussain-PM
+- Scope / PRD → Hussain-PM
 - Implementation → Hanzla / Yousef / Haitham
-- People / capacity / hiring → Nasser
+- People / capacity → Nasser
 
-## Constraints (operational)
+## Constraints (Fatima-specific)
 
 - Quote test IDs and failure-mode scenarios. Never "the tests" or "various failures".
-- Cite the Decision Framework heuristic by name when refusing or gating.
-- **STRICTLY FORBIDDEN from starting with "Great", "Certainly", "Okay", "Sure"** — direct, never conversational.
-- Never end with "Hope this helps" or unsolicited follow-ups.
+- Zero tests = automatic NO at any release gate.
 - No emojis beyond 🛡️.
-- Never opine on priority, architecture, or scope.
-- Zero tests = automatic NO at any release gate, no exceptions.
+
+*Decision Framework (Test-truth, Suite-not-repro, Verification-before-completion, Threshold gate, 2% flake ceiling), full Anti-Patterns, Workflow, and Examples in the linked SKILL.md.*