v2.1.0

v2.1.0 stabilizes the v2 session mesh and time subsystem for daily use, adds fuel (3-meter proprioception), and replaces CodeRabbit with a structured claude + codex review team. v3.0.0 (~6 weeks) will expose mine as observability-as-a-service via MCP and wire the repo to the claude code changelog within an hour of publish. see docs/rfcs/ for the shape.

what's new

cc plugin: time subsystem

• time rule at plugins/cc/rules/time.md: bimodal session modes (quick / standard / marathon), model x effort throughput matrix, 3-tier parallelism (main / subagent / agent teammate), relaxed compaction warnings for the 1M-context era
• SessionStart hook reads ~/.claude/mine.db and injects last-5-sessions-in-project timing. silent on missing db
• three skills: /cc:time-estimate, /cc:time-calibrate, /cc:time-benchmark. user-invocable: false so they stop cluttering the slash menu; still fully model-callable

cc plugin: slash ux

• plugin command renamed /cc:cc -> /cc:sessions (readable in the slash menu)
• every example in plugins/cc/README.md updated

fuel plugin (NEW, replaces the orphaned pulse@cc)

• 3-meter proprioception: 5-hour session, 7-day weekly, 200k conversation context
• pre-turn awareness hook injects threshold-aware context (80% soft warn, 90% suggest handoff, 95% dramatic intervention)
• /fuel state reads all meters; /fuel handoff drafts a clean stopping point + next-session opening prompt; /fuel quiet / /fuel loud toggle
• shipped briefly as pulse@cc v1.0.0, renamed to fuel for clearer gauge metaphor. migration note in plugins/fuel/README.md

review automation

• replaces CodeRabbit with an ai-review-team: claude-review (prose, voice, contract shape) + codex-review (bugs, security, shell/python/ci correctness). both curl-based against Anthropic / OpenAI responses apis. both read .github/AI_REVIEW_RUBRIC.md so output is consistent: blocking / apply / discuss / dismissed
• .coderabbit.yaml disables CodeRabbit auto-reviews; full uninstall via github repo settings -> integrations
• docs/review-process.md explains the contract, required secrets, how to turn reviewers off

auto-merge for autonomous PRs

• prs from @anipotts, trusted bots (dependabot, github-actions, pre-commit-ci, claude*, codex*), or autonomous branch namespaces (claude/, auto/, release/, chore/, fix/) get auto-merge enabled on open
• ci still gates the merge. red ci blocks the same way it always did
• trust list + failure modes documented at docs/auto-merge.md

translation without api keys

• translate.yml now falls back to argos-translate (fully offline pip package, CC-BY-4.0) when ANTHROPIC_API_KEY is unset
• argos helper at .github/scripts/translate_readme.py extracts prose between markdown-structural lines, translates only prose, reassembles. structural validation still runs
• claude-haiku path unchanged when key is set. commit message encodes which backend ran

docs + governance

• top-level README repositioning: subtitle anchored in where these patterns have been used (yc startups, public tech companies, unicorns). dropped star-CTA for "start here" links
• CONTRIBUTING.md, SECURITY.md, CODE_OF_CONDUCT.md
• three v3-horizon rfcs drafted

security + housekeeping

• hono dependabot alert #7 patched via transitive bump (4.12.12 -> 4.12.15)
• no u+2014 in any shipped markdown or JSON
• tested-with stamps refreshed to v2.1.118; version-stamps ci check now enforces (was informational)

breaking

• /cc:cc slash command no longer exists. use /cc:sessions. the underlying cc mcp tool is unchanged
• pulse@cc orphaned. the plugin is now fuel@cc. if you ran the old v1.0.0 pulse, rm ~/.claude/.pulse_cache ~/.claude/.pulse_quiet before first /fuel run

upgrading

claude plugin marketplace update cc
claude plugin install cc@cc --reinstall
claude plugin install mine@cc --reinstall
claude plugin install fuel@cc

v3.0.0 horizon (~6 weeks)

• mine v2 observability-as-a-service (MCP interface, billing method + feature stack + session-to-outcome mapping)
• freshness watcher with sub-hour SLA on the always-on operator machine
• tip content pass (voice + abstractable-pattern rewrites for ultrathink, session-length, fast-mode, prompt-caching)

full diff

v2.1.77...v2.1.0

thanks to CodeRabbit for the last cycle of feedback; the rubric that replaced it was shaped by what the noise looked like.