fix(coordd): heartbeat-freshness guard — never inject keystroke into actively-working agent

fix(mcp): drop sequential-thinking + postgres — Bash covers their value

fix(mcp): drop git MCP server — handshake unstable, git ops via Bash work

fix(mcp,codex): uvx→uv tool run for git MCP; strip mutable hooks.state from canon profiles

feat(codex): per-role profile-v2 setup + start_agent codex branch wires --profile-v2

refactor(plugins): delete 20 generic SKILL.md + 3 empty role plugins; rely on installed Anthropic/community plugins for stack-level skills

refactor: unified kebab-case across all role-name surfaces

Revert "fix(plugins,start_agent): drop tr-transform; plugin dirs use COORD_ROLE underscore form directly"

fix(plugins,start_agent): drop tr-transform; plugin dirs use COORD_ROLE underscore form directly

fix(plugins,start_agent): role plugin path derivation handles COORD_ROLE underscores

fix(stop_decide): skip processed-* inbox files so acked breadcrumbs don't trigger stop-block

fix(start_agent): -- separator before PROMPT to stop claude --mcp-config variadic from swallowing it

fix(plugins,lint): role-reader SKILL.md YAML + lint-tokens placeholder allowlist

docs: append canon skill plugin pointers to COORDINATE.md + every ROLE.md

feat(plugins): write all 38 canon SKILL.md files

feat(bin): wire layered plugins/MCP into start_agent + coord-init + lint-tokens

feat(plugins): canon plugin skeleton — 11 plugins + marketplace + canon MCP

docs: correct resume caveat — start_agent auto-resumes via .agent_registry

docs: single canonical fleet bring-up (daemon + tmux + maintainer-resume)

fix(start_agent): parse --mode loop|chat (was forwarded to tool as unknown flag)

fix(schema/bin): independent READER-audit path (A,B) + audit C/D/F holes