Skip to content

Latest commit

 

History

History
126 lines (87 loc) · 13.2 KB

File metadata and controls

126 lines (87 loc) · 13.2 KB

SunLit Orchestra — Claude Code overlay

This file is the Claude Code overlay for the canonical living catalog at docs/skill-pack-catalog.md. Use it when you are working in Claude Code and need Claude-specific session notes. For the host-neutral list of shipped skills, read the catalog first. For GitHub Copilot-specific notes, read copilot-instructions.md. For Codex-specific notes, read AGENTS.md.

Shared operating contract

Follow references/agent/operating-contract.md for the host-neutral agent rules: ask when ambiguous, make the smallest safe change, respect SLO allow-lists, verify before claiming, and keep host-boundary claims honest. Detailed workflows still live in docs/skill-pack-catalog.md and the installed skills/<name>/SKILL.md files.

Skill pack — first-party /slo-* skills

Sprint flow: Think → Plan → Build → Review → Test → Ship → Reflect.

Stage Skill Purpose
Ideate /slo-ideate YC-style product interrogation before any code
Research /slo-research Host-native research first; optional Claude batch backend via sldo-research
Architect /slo-architect Stack + ARCHITECTURE.md + interfaces lock-in + tla_required / kani_required flags
Verify design /slo-tla TLC model-check the design (when tla_required: true)
Verify code /slo-kani Kani model-check small bounded Rust kernels (when kani_required: true); code-level peer to /slo-tla
Plan /slo-plan Interactive v3 runbook authoring, one milestone at a time
Critique /slo-critique Four-persona adversarial review (CEO, eng, security, design)
Execute /slo-execute M<N> Per-milestone driver with allow-list enforcement
Verify /slo-verify M<N> Runtime QA with Playwright for UI surfaces
Close /slo-retro M<N> Lessons + completion + tracker update
Ship /slo-ship Open PR with runbook-aware description

Power tools:

Skill Purpose
/slo-second-opinion Cross-model disagreement surfacer (Codex / Gemini)
/slo-freeze <path> Lock edits to one directory for the session
/slo-resume Read current runbook's tracker, suggest next step

Innovation Sandbox loop — first-party /slo-experiment discovery lane

A pre-idea discovery lane that sits before the Sprint flow and feeds it (promote_to_idea/slo-ideate, etc.). It turns a fuzzy technical hunch into a promotable candidate or a documented dead-end, authored into one durable artifact — the Experiment Book (docs/slo/experiments/<slug>/EXPERIMENT.md), the experimentation peer of the v4 runbook: Definition of Learned, not Done. Canonical list: docs/skill-pack-catalog.md; loop: docs/LOOPS-ENGINEERING.md; design: docs/slo/design/innovation-loop-overview.md. M1 ships /slo-experiment + the template; the 7 phase skills (/slo-sandbox/slo-demo) land M2–M5 of docs/RUNBOOK-innovation-loop.md. Spike scratch lives under experiments/<slug>/<spike-id>/ (git-ignored); nothing promotes to production without re-entering the Sprint or Ticket loop.

Biz skill pack — first-party advisor + generator skills

UK-only (v1). Advisor skills (4) operate as draft | translate | triage | prepare modes with hard-block gates from references/biz/triage-gate.md (regulated / >£5,000 / counterparty-with-lawyer / GDPR). Generator skills (11; shipped in Runbooks B1, B2, C — out of Runbook A scope) produce one artifact each.

Skill Archetype Domain Cites
/slo-legal advisor UK legal — NDA, contractor SOW, IP assignment, T&Cs references/biz/templates/onenda-uk.md (CC BY-ND 4.0 verbatim), references/biz/cost-baseline-jpp-law-2026.md
/slo-accounting advisor UK accounting — bookkeeping, VAT, R&D credit, MTD HMRC route default = accountant (per references/biz/jurisdiction-uk.md UK regulator index)
/slo-equity advisor UK equity — cofounder split, vesting, cap-table snapshot references/biz/hmrc-vcm-index.md (VCM34080 / VCM3000 / VCM31000 + Abingdon Health line) — runs SEIS / EIS pre-check on every draft
/slo-fundraise advisor UK fundraise — SAFE math, pitch narrative, term-sheet redline brief HMRC VCM index + references/biz/ir35-cest-factors.md — runs Advance Assurance pre-check on every interaction; refuses term-sheet drafting without AA ≥ 6 weeks ahead
/slo-talk-to-users generator UK user-interview prep + post-interview extraction (Mom Test discipline); single-mode with mode_arg: pre-interview | post-interview; output docs/biz/users/<date>-<name>.md (confidential) Lands /slo-verify Pass 4 PII-pattern scan over docs/biz-public/ (Runbook B1 M1)
/slo-gtm generator UK GTM strategy — ICP / segmentation (3-segment cap) / motion choice (PLG | sales-led | community-led | hybrid) / channel strategy / KPI alignment; output docs/biz-public/gtm/strategy.md Routes direct-marketing channels to /slo-legal triage for PECR considerations
/slo-product generator UK PM artifacts (mode_arg roadmap | metrics | okrs); PM-side metrics only (DAU / activation / retention / feature adoption); 3-objective OKR cap Redirects financial KPIs (CAC / LTV / NDR / burn multiple) to /slo-metrics (Runbook B2)
/slo-marketing generator UK marketing tactics (mode_arg b2b | b2c); brand voice / content calendar / channel mix / demand gen / paid acq Routes ALL direct-marketing implementation to /slo-legal triage for DUAA 2025 PECR (£17.5M ceiling); B2C flags ASA / CAP Code disclosure + CRA 2015
/slo-launch generator UK launch sequence — pitch validator + 4-stage launch (silent → F&F → communities → press) with kill / delay rules per stage; output docs/biz-public/launch-<slug>.md Routes any direct-marketing implementation in stage 4 to /slo-legal triage for PECR; readiness checklist gates broader-press launch
/slo-sales-funnel generator UK outbound funnel math + cold-email template (7 outbound principles) + deal structure (paid trial → recurring → opt-out); output docs/biz-public/sales/funnel-<segment>.md Routes cold email to /slo-legal triage for PECR (gate-4-gdpr-document fires under DUAA Stage 3)
/slo-pricing generator UK pricing strategy — value equation (price = 25-33% of value delivered) + 3-tier-max model + canonical "increase price by 50%" experiment framing; output docs/biz-public/pricing.md Routes SEIS/EIS revenue-mix considerations to /slo-fundraise triage for VCM3000 qualifying-trade check
/slo-metrics generator UK financial KPI dashboard (mode_arg consumer | b2b); CAC / LTV / NDR / MoM revenue / burn multiple / gross margin / runway / ARR; output docs/biz-public/metrics.md Redirects PM-side metrics (DAU / activation / retention / feature-adoption) to /slo-product metrics (Runbook B1 M3)
/slo-cofounder generator UK cofounder evaluation + 4-week paid trial framing + monthly 1:1 agenda; output docs/biz/cofounder/<name>.md (confidential — real persons) Routes equity-split to /slo-equity (A M3); routes active disputes to a mediator + /slo-legal triage
/slo-hire generator UK hiring artifact with mode_arg = role-shape swe | ae | designer | ops; sourcing + interview rubric + offer cadence + onboarding; output docs/biz/hires/<role>-<name>.md (confidential) MANDATORY IR35 triage gate per references/biz/ir35-cest-factors.md — runs the seven hard-block triggers BEFORE offer; rejects "call them a contractor for tax efficiency" framing
/slo-founder-check generator UK founder self-assessment — 12-question check + worst-case-runway worksheet (4 cost-cut tiers) + optional YC application prep; output docs/biz/founder-check.md (confidential — highly personal) Routes mental-health to professional support; routes personal-finance to StepChange / IFA; routes wind-down questions to /slo-legal triage

Shared scaffolding lives at references/biz/ at the repo root (NOT under skills/crates/sldo-install/src/install.rs:44-71's discover_skills() ignores it). Two-tier output: docs/biz/ (gitignored, confidential drafts with real PII / deal terms) and docs/biz-public/ (git-tracked, placeholder / decision artifacts). See docs/slo/design/biz-skill-pack-overview.md for the full design.

GDPR posture: broad hard-block on draft for all GDPR documents (privacy notice, ROPA, DPA, internal policies). Locked 2026-04-25. Reversal requires fresh /slo-architect pass. Cost baseline: JPP Law fixed-fee public pricing, locked 2026-04-25.

PII discipline: every biz skill writing to docs/biz/ issues a write-time warning when the target dir is git-tracked AND a remote exists AND tier: confidential. Second-line defense: /slo-verify Pass 4 PII-pattern scan over docs/biz-public/ (added Runbook B1 M1) — flags email / UK NI / sort code / capitalised-bigram patterns with pii_scan_override: true + tier_override_reason: <one-liner> frontmatter override mechanism.

Third-party skills vendored

Skill Purpose Prereq
/get-api-docs Fetch current third-party API docs via chub CLI npm install -g @aisuite/chub

See skills/get-api-docs/UPSTREAM.md for attribution.

Specialist agents (optional, Claude-only — sap-imp M5)

Four host-native specialist agents under agents/ provide an additive Claude-Code-only critique flow: slo-runbook-review-lead, slo-security-reviewer, slo-design-reviewer, slo-verification-lead. Output paths are constrained to docs/slo/critique/ and docs/slo/verify/ (same paths /slo-critique and /slo-verify write to). The structural-contract test in xtasks/sast-verify/tests/sap_imp_m5_agents.rs enforces frontmatter, output-path safety, and /slo-critique SKILL.md SHA-256 byte-identical baseline.

/slo-critique persona rotation remains the canonical portable critique path. Agents are an optional enhancement; the canonical flow is preserved on every host.

Examples gallery

Synthetic, non-normative gallery at examples/. Read these to see what shipped SLO outputs look like before running a skill.

Distribution channels

  • sldo-install (canonical, multi-host) - installs into the selected host root.
  • .claude-plugin/plugin.json (optional, additive) - Claude Code organizational installs may prefer a one-zip distribution. Tagged releases produce a downloadable zip via the SHA-pinned release-zip workflow.

Canonical planning artifact

Every new feature runbook lives at docs/RUNBOOK-<FEATURE>.md and follows docs/slo/templates/runbook-template_v_4_template.md. This is the v4 template — the output contract of /slo-plan — which adds Carmack-style reliability controls (debugger-first inspection, mandatory static analysis, assertion-driven invariants, bounded resource design, "make invalid states unrepresentable") on top of v3's SunLit-specific structure (carry-forward from prior retros, abuse-acceptance scenarios, threat-model integration). Do not bypass it for batch CLI shortcuts when interactive planning is an option. docs/slo/templates/runbook-template_v_3_template.md remains in place as the historical artifact for runbooks already authored against v3.

Baseline test command (this repo)

cargo test -p sldo-common -p sldo-install -p sldo-research

The workspace contains four crates: sldo-common (shared library), sldo-research (optional Claude batch backend for /slo-research), sldo-install (skill installer), and xtasks/sast-verify (Semgrep rule gate driven by /slo-rulegen + /slo-ruleverify). All other Rust code (the legacy sldo-plan / sldo-run CLIs, the parked sldo-tauri desktop UI, the sldo-tla-sha maintenance utility) was removed in the 2026-04 cleanup — the skills are the canonical interface now.

Installing the pack on this machine

The fastest path is cargo install from crates.io:

cargo install sldo-install
sldo-install                # global: ~/.claude/skills/
sldo-install --dry-run      # preview
sldo-install uninstall      # reverse

From a checkout (when iterating on the skill pack itself):

cargo build -p sldo-install --release
./target/release/sldo-install              # global: ~/.claude/skills/
./target/release/sldo-install --dry-run    # preview
./target/release/sldo-install uninstall    # reverse

Manifest: ~/.sldo/install.toml.

If you are using GitHub Copilot instead of Claude Code, use copilot-instructions.md for the matching host overlay. If you are using Codex, use AGENTS.md.

If you are completely new to the repo, start with docs/getting-started.md before using this overlay.