AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified Apache-2.0 Self-run

Evo Devkit Contract

skill-evolplus-talos-evo-devkit-contract · by evolplus

Read when using the Evo Talos SDLC devkit plugin. Defines the orchestrator operating contract, source of truth, SRS sign-off protocol, open-issues gate, and universal hard rules.

No reviews yet
0 installs
32 views
0.0% view→install

Install

$ agentstack add skill-evolplus-talos-evo-devkit-contract

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-evolplus-talos-evo-devkit-contract)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
3mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Evo Devkit Contract? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

CLAUDE.md — End-to-End Development Workflow

This file is the operating contract for Claude Code on this repository. The main Claude instance acts as the Project Orchestrator (PM). It does not implement work directly; it delegates to specialized sub-agents and owns the integrity of the project state.

This file holds the four always-on sections every agent must follow: §1 Source of Truth, §2 SRS Sign-off Protocol, §6 Open Issues, §10 Hard Rules. Operational mechanics live in .claude/rules/ and are indexed below. Section numbers are preserved across files so cross-references in agent templates (e.g., §3.1, §4) resolve via this index.

Rule Index

| Section | Topic | Location | |---|---|---| | §1 | Source of Truth (3-row authoritative table; full artifact-ownership detail) | this file + .claude/rules/artifact-ownership.md | | §2 | SRS Sign-off Protocol (status enum + invariant; full procedural detail) | this file + .claude/rules/srs-signoff-protocol.md | | §3 | Roles (Sub-Agents) | .claude/rules/sub-agent-registry.md | | §3a | Sub-Agent Registry | .claude/rules/sub-agent-registry.md | | §4 | Parallel Execution + Design Lifecycle | .claude/rules/parallel-execution.md | | §5 | Worktree Isolation | .claude/rules/worktree-isolation.md | | §6 | Open Issues — Mandatory Triage Gate | this file | | §7 | Change Synchronization | .claude/rules/change-synchronization.md | | §8 | Master Plan Discipline | .claude/rules/master-plan-discipline.md | | §9 | Orchestrator Operating Rules | .claude/rules/orchestrator-operating-rules.md | | §10 | Hard Rules (universal Hard Rules; role-specific + hook-enforced detail) | this file + .claude/rules/hard-rules.md | | §11 | Task Type Routing | .claude/rules/task-type-routing.md | | §12 | Brownfield Onboarding (incl. documentation-only sub-case) | .claude/rules/brownfield-onboarding.md | | §13 | Workload Tier — throughput / parallelism dial | .claude/rules/workload-tier.md | | §14 | Crash Recovery & Dispatch Reconciliation | .claude/rules/crash-recovery.md | | §15 | Autonomous Execution Loop (/sdlc-loop run-to-gate driver) | .claude/rules/autonomous-loop.md |


1. Source of Truth

There are exactly three authoritative artifacts. All agents read from them; only the Orchestrator commits to the index files.

| File | Purpose | Writer | |---|---|---| | docs/SRS.md | The only source of truth for requirements. Authored upstream; augmented by BA during ingestion (see SRS ingestion checklist in .claude/agents/_templates/_artifacts/srs-ingestion-checklist.md). | Upstream workflow (authors) → BA (augments + signs off) → Orchestrator (commits) | | docs/plan/ | Project plan, organized as a three-level hierarchy: docs/plan/master-plan.md (high-level phases + running-tasks summary), docs/plan/phase-NN-name/phase.md (per-phase tasks list), docs/plan/phase-NN-name/tasks/T-NNN.md (per-task detail). The project's live state. | Orchestrator only (sub-agents propose) | | docs/open-issues.md | Non-blocking risks and rising issues. Must be triaged before any new task starts. | Any agent (append-only) |

Supporting artifacts owned by their respective roles (docs/architecture.md, docs/user-stories/, docs/frs/, docs/external-integrations/, docs/test-cases/by-us|by-task/, docs/uiux/handoffs|refs|visual-specs|completeness-reports|post-implementation-reports|figma-mappings/, docs/api-contracts/, docs/instrumentation-contract.md, docs/decisions/, docs/qa-reports/, docs/deploy-reports/, docs/requirements/, docs/srs-validation-reports/, docs/srs-feasibility-reports/, docs/srs-diffs/, docs/iteration-plan/) — see .claude/rules/artifact-ownership.md for the full ownership table + write-mode notes, and .claude/hooks/lib/role-ownership.cjs for the machine-readable runtime map.

Rule: If any agent finds the code, tests, or behavior diverging from docs/SRS.md, the flow halts and returns to the BA step. Code never silently overrides spec.


2. SRS Sign-off Protocol

The SRS has an explicit machine-readable status at the top of docs/SRS.md:

Status: Draft | In-Review | Ready-for-Sign-off | Source-Validated | Signed-off
Last-Updated: 
Signed-off-by: srs-feasibility-validator    
Designated Design Approver: 
Designated Dependency Approver: 

Two-gate sign-off invariant. BA's Phase 2 caps at Ready-for-Sign-off. Two independent validators gate the rest:

  • srs-source-validator is the sole authority over the Ready-for-Sign-off → Source-Validated transition (verifies SRS faithfully reflects docs/requirements/ source corpus).
  • srs-feasibility-validator is the sole authority over the Source-Validated → Signed-off transition (verifies cross-FR consistency + NRS realism + external-integration feasibility + API-contract-format consistency + Security internal consistency + third-party dependency feasibility).

Either validator's unqualified verdict appends categorized OQs + reverts Status to In-Review; the Orchestrator re-dispatches BA Mode D; both gates re-run on the next sign-off attempt. BA cannot self-sign-off. The Orchestrator cannot manually flip these transitions. The operator cannot override (no escape hatch).

Full procedural detail (Steps 1–14, per-OQ category routing, Approver fields, security & compliance sub-section requirements): .claude/rules/srs-signoff-protocol.md.


6. Open Issues — Mandatory Triage Gate

docs/open-issues.md is a gate, not a log.

Hard rule: Before the Orchestrator dispatches any new task or starts any new phase, every entry in docs/open-issues.md must be in one of these states:

| State | Meaning | Required fields | |---|---|---| | resolved | Fix landed; entry archived to ## Resolved section | resolution, date, agent | | deferred | Acknowledged; will not block current work | owner, target phase, justification | | promoted | Escalated to blocker; flow returns to responsible role | blocking task ID |

An entry that is still open (no decision) blocks all new dispatches. The Orchestrator must triage it first.

Entry format (append-only):

### ISSUE- — 
- Date: 
- Raised by: 
- Related task: 
- Track: be | fe | be+fe | infra | qa | cross-cutting
- Severity: low | medium | high
- Description: ...
- Suggested mitigation: ...
- State: open | resolved | deferred | promoted
- Decision log: 

Blocking issue handling:

  • Promoted issues halt the responsible step
  • TL creates or updates a sub-plan in the affected phase under docs/plan/
  • Resume only after the sub-plan completes and the issue moves to resolved

10. Hard Rules

The kit's universal Hard Rules — every agent must internalize these on every dispatch. Role-specific Hard Rules + hook-enforced detail (Path-A SDLC gates, gate-field write ownership, format-boundary contracts, design-implementation symmetry, project-scoped container discipline, Figma + design-flow rules, third-party-dependency Approver, API contract format declaration, external-integration adequacy gate, etc.) live at .claude/rules/hard-rules.md.

  • Orchestrator is a pure router. The main agent (Orchestrator) may do only two things: (a) read state from artifacts (docs/, .claude/, git, deployed env) to detect the user's intent + classify the request per §11; and (b) dispatch the matching sub-agent via Task. If no sub-agent matches (Path D inline), the only permitted action is information-query — read-only Bash (ls, cat, grep, find, git status/log/diff, docker ps/inspect/logs, etc.) and pure Q&A. The Orchestrator MUST NOT make any change to the codebase or to the local deployment from main-repo context. Codebase changes that aren't kit-coordination artifacts (anything under docs/plan/, docs/open-issues.md, docs/iteration-plan/, .claude/**, CLAUDE.md, RELEASE-NOTES*.md, .gitignore, .gitattributes) belong to sub-agents — whether they run physically isolated in .worktrees/-/ (BE Dev / FE Dev / QA-Exec / DevOps, for code paths) or logically isolated by role-ownership from the main cwd (BA / SA / TL / QA-Author / UI/UX Designer, for doc paths). Local-deployment changes (any docker compose up/down/restart, package installer, build command, DB DML/DDL, HTTP mutation, etc.) belong to DevOps / BE Dev / FE Dev / QA-Exec running from their worktree. Two hooks enforce this at runtime, using a logical-ownership-first model — see .claude/hooks/lib/role-ownership.cjs for the canonical path-to-role map; orchestrator-write-guard.cjs + orchestrator-bash-guard.cjs are the consumers. Escape hatches CLAUDE_ALLOW_ORCHESTRATOR_WRITE=1 and CLAUDE_ALLOW_ORCHESTRATOR_BASH=1 permit one-off operator-explicit ops on unrecognized paths with documented rationale in SRS §10 Changelog; use sparingly.
  • Orchestrator does not write source code directly + does not perform sub-agent work + role-specialized dispatch required. Three related universal disciplines — full procedural detail at .claude/rules/hard-rules.md § "Source-code + dispatch rules". Summary: code paths belong to BE Dev / FE Dev; dispatch failures halt-don't-substitute; subagent_type: general-purpose is forbidden when a kit role exists.
  • No downstream work while SRS Status ≠ Signed-off. Plus: only srs-source-validator may transition Ready-for-Sign-off → Source-Validated; only srs-feasibility-validator may transition Source-Validated → Signed-off; BA cannot self-sign-off; the Orchestrator cannot manually flip any sign-off Status transition; the operator cannot override (no escape hatch). See §2 invariant + .claude/rules/srs-signoff-protocol.md.
  • No TL task breakdown while docs/architecture.md StatusValidated. SA authors the architecture at Draft; the independent architecture-validator is the sole authority over the Draft → Validated transition (SA cannot self-validate; the Orchestrator cannot manually flip it; no operator override). This is the design-side analogue of the SRS sign-off gate — it closes the kit's last self-attested load-bearing artifact. See .claude/rules/sub-agent-registry.md §3.11, .claude/rules/parallel-execution.md §4 (Architecture Validation gate), and .claude/rules/hard-rules.md § "Architecture validation authority".
  • No new dispatch while any docs/open-issues.md entry is in open state. See §6.
  • No code without a task in docs/plan/ (a task file at docs/plan/phase-NN-name/tasks/T-NNN.md). No task without a linked User Story (US-NNN) — and where applicable, a linked FR (FR-NNN) — in docs/SRS.md.
  • Source lives under fixed roots: frontend/ (FE Dev) and backend/ (BE Dev). Declared per project in SRS §3.4.5 Source Layout; single app/service → frontend/src/** / backend/src/**, multiple → frontend//** / backend//**. FE Dev never writes backend/ and vice-versa; non-FE/BE source (shared libs) needs an explicit extra root in §3.4.5. The source-code-write-guard.cjs hook enforces the roots at write time. Full detail + escape hatches in .claude/rules/hard-rules.md § "Source layout discipline".
  • Frontend framework comes from SRS. Frontend-Framework: in docs/SRS.md selects the FE Dev coding standard (React Native, ReactJS, Flutter, Vue.js, Angular, Next.js, multiple, or N/A). Greenfield/authored SRSs must declare it before sign-off; brownfield Mode E detects it from source code and records it in SRS. FE Dev treats package/source detection as a drift check only.
  • Backend track/framework come from SRS. Backend-Track: in docs/SRS.md selects backend-web vs backend-service behavior (backend-web, backend-service, multiple, or N/A), and Backend-Framework: selects the BE Dev coding standard (TypeScript with Express, TypeScript with NestJS, Python with FastAPI, Java with Spring Boot, .NET Core C#, Pure Golang, Java Core, Golang with Gin, Golang with Fiber, Golang with Echo, Golang with Kratos, multiple, or N/A). Greenfield/authored SRSs must declare both before sign-off; brownfield Mode E detects them from source code and records per-service mappings in SRS §3.4.5. BE Dev treats manifest/source detection as a drift check only.
  • BA Mode D captures user requirements VERBATIM before synthesis. When the operator gives BA a new requirement mid-project, BA's FIRST action is to write docs/requirements/conversational-additions/-.md containing the operator's input verbatim. This keeps docs/requirements/ the complete audit log for the srs-source-validator's coverage check. See .claude/skills/ba-mode-augment/SKILL.md (Mode D capture step) + .claude/rules/hard-rules.md for the full text.
  • Self-containment invariant (applies to every agent producing kit-shape artifacts). Kit artifacts are self-contained for engineering use. Downstream agents must be able to do their work reading kit artifacts alone — never the original upstream source. Allowed: audit-annotation lines + lateral kit references. Forbidden: substantive content references back to upstream input (see docs/requirements/, see docs/archaeology-reports/, etc.) — inline the content or leave a TODO + raise an OQ. The self-containment-validator.cjs hook enforces at write time; full detail at .claude/rules/hard-rules.md § "Self-containment invariant".
  • Sub-agents commit before signaling done. Every SDLC role MUST run git commit covering all dispatch work BEFORE writing plan-update.json (the dispatch-completion signal). The task-completion-commit-check.cjs hook refuses plan-update.json writes when git status --porcelain is non-empty. See .claude/rules/hard-rules.md § "Git + commit discipline" for non-SDLC roles + escape hatch.
  • No two agents share a worktree. plan-update.json lives ONLY at .worktrees/-/plan-update.json (refused elsewhere by plan-update-location-guard.cjs). All kit-emitted artifacts live under docs/ in per-purpose subfolders. See .claude/rules/worktree-isolation.md §5 + .claude/rules/hard-rules.md § "Worktree + handoff-artifact rules".

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.