v0.1.1

Fixed

• templates/develop/SPRINTS.md.tmpl shipped with damaged encoding (BOM + mojibake
  em-dashes); a literal render produced corrupted headers. Re-encoded clean UTF-8. Found by
  replaying the public walkthrough against the installed copy.
• templates/PLACEHOLDERS.md declared a {{BRANCH_PREFIX}} default that disagreed with the
  loop skill's (claude/… vs cdev/sprint-<nn>-batch-<nn>) — two shipped documents
  contradicting each other about the same switch, the exact incident class the method's own
  consistency rule exists for. Unified on the loop's default.

Added

• docs/community/walkthrough.md — replay-validated first-use walkthrough.
• docs/community/rfc-process.md — the RFC process with its field-evidence bar.
• Community layer: CONTRIBUTING, CODE_OF_CONDUCT, SECURITY, issue forms (field report
  included), PR template, labels.
• Visual identity (assets/): master icon, flat vector marks, wordmark lockups, social
  preview.
• Public README: positioning, the four failure modes, the loop, quick start, command table,
  lifecycle guide, evidence tables, and the multi-repo/microservices section with the
  coordination-flow diagram.
• docs/09-cdev-monorepo.md — the multi-repository layer explained, with the flow diagram
  and a backend → frontend worked example.
• docs/10-usage-recommendations.md — model/effort defaults for the current Claude Code
  binding (labeled as binding guidance, dated).
• First CI: .github/workflows/validate.yml runs scripts/validate.sh (portable bash) on a
  Linux/Windows/macOS matrix — manifest parses, zero unresolved placeholders in consumed
  files, skill frontmatter contract, internal links resolve (#3).

Changed

• The repository guide is now AGENTS.md — the agent-neutral standard read by Codex,
  Cursor, Jules, Zed and others. Hosts that read a different filename get a pointer file
  (Claude Code: a CLAUDE.md containing @AGENTS.md). Conditioning renders both; the
  skills and the verifier read the guide by its neutral name. First step of making
  conditioned repositories usable by any coding agent.

Removed

• Internal working material (launch drafts, strategy notes, stale marketplace draft) moved
  out of the public tree.

CDev v0.1.0 — first public release

What CDev is

Continuous Development Framework for Coding Agents. Coding agents work in sessions; CDev turns those sessions into continuous development: the repository — not the conversation — holds the plan (SPRINTS.md), the state (AGENT_PROGRESS.md) and the proof (git + recorded evidence). Any session resumes by reading, never by remembering.

What this release includes

• Seven skills, namespaced cdev:*:
  • /cdev:bootstrap — condition one repository (recognition-first, structure + clarity halves, single human gate)
  • /cdev:cdev — the execution loop (dispatcher + one loop reading role profiles)
  • /cdev:cdev-planner — materialize an objective into batches, or gap-analyze
  • /cdev:ockham — re-tell dense technical output in plain language (user-invoked only)
  • /cdev:bootstrap-monorepo, /cdev:cdev-monorepo-planner, /cdev:cdev-monorepo — the optional multi-repo orchestration layer
• Role profiles (profiles/): backend and frontend — each exactly four things: what verified means, what evidence means, the hardest gate, the self-chosen work order.
• Conditioning templates (templates/) with a placeholder contract.
• The method's documentation (docs/01–08): origin, model, mechanics, skills, two field reports, core-vs-periphery, installation.

Current binding

Designed as a coding-agent-independent method. Currently shipped as a Claude Code plugin — the first host binding.

Install

/plugin marketplace add elmayii/cdev-marketplace
/plugin install cdev@cdev-marketplace

Known limitations

• One host binding (Claude Code); no other agent validated yet.
• Field evidence comes from one product — deep (six weeks, five CDev instances, four repository kinds), but one. Second-product validation is the next phase.
• Support scripts are PowerShell.
• No fullstack profile — deliberately, until something real runs on one.

Experimental

The multi-repo orchestration layer ran nine days in production and broke in instructive, documented ways (docs/06); its five field-derived rules are new in this release and not yet re-validated in the field.