v0.106.0

Minor Changes

• 375922a: Continue block directories are now walked recursively, matching Continue's own loader: MCP block files in nested subdirectories of ~/.continue/mcpServers/ (and project .continue/mcpServers/) plus nested prompts/rules YAML block files are discovered in sorted relative-path order. Single-server JSON files keep their basename-derived names and duplicate handling is unchanged.

v0.105.0

Minor Changes

• 97fe2a9: Continue JSON MCP block files now support the full upstream loader matrix: JSONC comments, claude-code style files with projects nesting, and single-server files (server name = filename), in both ~/.continue/mcpServers/ and project .continue/mcpServers/. envFile on stdio entries emits a not-migrated warning and unsupported file shapes are skipped with an explicit warning instead of being silently ignored.

v0.104.0

Minor Changes

• b8972c9: Continue user-level MCP block files: servers defined in ~/.continue/mcpServers/ (YAML files with an mcpServers: list plus claude-style JSON maps) are now exported alongside the config.yaml list, matching Continue's own loading. config.yaml entries win duplicate names with a per-entry warning, and the project adapter's .continue/mcpServers/ reads share the same parser and duplicate handling. Imports are unchanged and never write the block files.

v0.103.0

Minor Changes

• bf5bc1a: Continue inline prompts and rules: prompts:/rules: entries defined inline in ~/.continue/config.yaml and in .continue/prompts/*.yaml / .continue/rules/*.yaml local block files (user and project scope) are now exported. Inline prompts become synthesized markdown prompts with invokable: true frontmatter (matching continue's own prompt template; markdown prompt files win name duplicates with a warning); inline rules are merged into the exported instructions document, with a warning when scoping metadata (globs, regex, alwaysApply, invokable) is dropped. Hub uses: block references and malformed entries are skipped with warnings. Imports are unchanged and never write inline config entries.

v0.102.0

Minor Changes

• OpenCode inline config entries: agents, commands and primary modes defined inline under the agent, command and mode keys of opencode.json/opencode.jsonc (in ~/.config/opencode/ and the ~/.opencode/ fallback; project scope reads the root and .opencode/ config files) are now exported as synthesized markdown, matching opencode's merge order: the inline mode map merges last and wins duplicates with mode: "primary", while markdown files beat inline agent/command entries from the same config dir. Each exported inline entry gets a warning naming its source file; disable: true entries and inline commands without a template are skipped with warnings, and {file:...}/{env:...} placeholders are copied as-is with a warning. Imports are unchanged and never write inline config entries. (#249)

npm: https://www.npmjs.com/package/agentmove-cli/v/0.102.0

v0.101.0

Minor Changes

• 6d001ad: OpenCode primary modes now migrate: {mode,modes}/*.md in both ~/.config/opencode/ and the ~/.opencode/ fallback config dir (project: .opencode/{mode,modes}/) are exported into the custom agents layer byte-faithfully. Matching opencode's merge order, the scan is flat (nested mode files are ignored) and a mode beats a same-name agent within its config dir; each mode entry carries a warning noting opencode loads it with mode: "primary". Imports still write only the native agents roots and never synthesize mode files.

v0.100.0

Minor Changes

• a89b3cf: OpenCode custom agents and commands now merge every root opencode actually scans: {agent,agents}/ and {command,commands}/ in both ~/.config/opencode/ and the ~/.opencode/ fallback config dir (project: .opencode/{agents,agent}/ + .opencode/{commands,command}/), read recursively with nested names preserved. Duplicate names keep the fallback-dir copy (matching opencode's last-config-dir-wins merge) with an explicit shadow warning; imports still write only the native plural roots.

v0.99.0

Minor Changes

• c66a33e: OpenCode multi-root Agent Skills: exports now merge ~/.config/opencode/{skills,skill}/, the ~/.opencode/{skills,skill}/ fallback config dir, and the generic shared root ~/.agents/skills/ (project: .opencode/{skills,skill}/ + .agents/skills/), matching opencode's own skill discovery. On duplicate names the first root in that order wins with a shadow warning; imports keep writing only ~/.config/opencode/skills/.

v0.98.0: kimi generic ~/.agents/skills root

agentmove-cli 0.98.0

Kimi Code CLI: generic shared skills root (#241, #242)

Kimi Code CLI loads Agent Skills from both its brand root and the generic shared Agent Skills root. AgentMove now matches upstream (MoonshotAI/kimi-code packages/agent-core/src/skill/scanner.ts):

• Export merges ~/.kimi-code/skills/ + ~/.agents/skills/ (project: .kimi-code/skills/ + .agents/skills/ with --project).
• On duplicate names (compared case-insensitively, matching kimi's normalizeSkillName) the brand root wins with a shadow warning — kimi's scanner registers first-wins in brand-then-generic order.
• Import keeps writing only the brand root (.kimi-code/skills/), which has top scanner priority, avoiding double-ownership of the shared .agents/skills root used by other clients.
• Honestly deferred (documented in docs/gap/GAP-ROUND-129.md): kimi's flat single-file <name>.md skills, $KIMI_CODE_HOME relocations, extra_skill_dirs/--skills-dir, plugin/builtin skill roots, sub-skill qualification.

Docs: https://agentmove.zalize.com/docs/clients/ · Install: npm i -g agentmove-cli

v0.97.0

Minor Changes

• eea5c6b: Grok CLI: custom sub-agents interop + real skills root. Export converts subAgents entries from ~/.grok/user-settings.json into portable agents (instruction only; the per-agent model is dropped with a warning; reserved built-in names, duplicates, and instruction-less entries are warned and not migrated). Import merges portable agents back into subAgents with grok's default model, preserving all other user-settings keys and existing entries (same-name overwrites, nested-name flattening, and reserved-name skips are warned). Agent Skills now read/write grok's real root ~/.agents/skills (project .agents/skills); the legacy ~/.grok/skills root — which grok never loads — is still read on export with warnings.