monomind-release-rules
Standing rules of monomind's release org, pinned to every role: names, command prefix, scratch and git limits, evidence close, hand-off documents.
- License 许可
- Apache-2.0
- 所属仓库
- monoes/monomind
- 最近更新
- 2026-11-10
- Stars
- ★ 23
- Forks
- 2
01 前言 Front Matter
- name
- monomind-release-rules
- tags
- [ "operations", "devops" ]
- tools
- []
- source
- https://github.com/monoes/monomind
- license
- Apache-2.0
- description
- Standing rules of monomind's release org, pinned to every role: names, command prefix, scratch and git limits, evidence close, hand-off documents.
03 SKILL.md 正文
Release rules (every role, every command)
Names. ORG_ROOT = your starting cwd (the owner's checkout; never change it). SRC, DOCS, FIX2 = ORG_ROOT/.monomind/orgs/release/work/{src,docs,fix-2}. GATE = $HOME/monomind-release/<VERSION>-<UTC ts>; TMPDIR = $HOME/mrg-tmp. VERSION, SHA, GATE and the ABSOLUTE worktree path come only from the task brief, never from an earlier run. Start a task with org_recall lesson <area> for each area the brief names; apply what fits, name it in your result.
Commands
- Prefix every node/pnpm/npm/vitest/monomind command:
env -u MONOMIND_SDK_AGENT -u MONOMIND_HOOK_QUIET -u MONOMIND_GRAPH_GATE -u MONOMIND_NO_LOCAL_EMBEDDINGS MONOMIND_CRASH_REPORTING=off MONOMIND_AUTO_UPDATE=false CI=true TMPDIR=$HOME/mrg-tmp npm_config_cache=$GATE/npm-cache(org sessions inject those four variables and change monomind's behaviour, #249). - Long work runs in the FOREGROUND, Bash
timeout: 600000, output to$GATE/logs/<area>/<step>.logthenecho "exit=$?" >> <log>; split anything over 10 min. Neverrun_in_background, never end a turn to wait (nothing wakes you; wait in anuntilloop). Keep turns short: org mail arrives only when a turn ends. Nothing outlives its Bash call; kill only PIDs you started, never pkill/killall/cleanup --force. - Read in bulk (one Bash call printing every range,
git show <sha> -- <path>,grep -n, Read offset/limit); never one-range peeks, whole logs or whole diffs. - Never write under /tmp. Never empty or recursively delete
$HOME/mrg-tmp(#273: it bricked a role's shell): delete only subdirectories you created. Never delete another run's GATE (earlier issues cite its logs). - Destructive commands (
cleanup,init --force, recursive deletes,git clean,git reset --hard,git checkout -- <path>) only in a scratch project under$HOME/mrg-tmp:cd $HOME/mrg-tmp/<check>-<short-sha> && case "$PWD" in $HOME/mrg-tmp/*) ;; *) exit 99;; esac && <cmd>. From ORG_ROOT,cleanup --forceonce deleted 1003 tracked files and the memory store. Damage outside scratch: stop, tell release-captain what ran and changed. - Never run
scripts/check-published-pins.mjsorprepublishOnlyoutside PUBLISH (waits ~8.5 min for versions that cannot exist yet).
Git. Never git config in any checkout of this repo (worktrees share it, #250); scratch repos: git -c user.name=qa -c user.email=qa@example.invalid. Nobody edits, commits in, checks out, deletes or moves anything in ORG_ROOT (sole exception: publisher's LOCAL MAIN SYNC); list strays in the report. Below policy.git push you have no GitHub credentials and at read .git is read-only: expected, not environmental (#258). Ask publisher for anything needing GitHub auth or a ref change; read public GitHub with unauthenticated curl https://api.github.com/repos/monoes/monomind/....
Evidence. Only this run's $GATE/logs count (mtime after run start, SHA inside the log; org logs times are UTC, #253). Copy any drill or scratch artifact you cite into $GATE/logs/<round>/ before cleaning up (#350). Per check: {name, command, exit, PASS|FAIL|SKIP|FLAKY, log, one line}; SKIP needs the exact error; never call a failure "environmental" without a reproduction.
Closing a task. org_task_done: result = check table + summary; evidence is its own argument {headSha, worktree, checks}, never text in result. worktree = the ABSOLUTE path of the worktree the checks ran in (SRC, DOCS or FIX2; omit only for ORG_ROOT), never a label like src; headSha = its CURRENT HEAD (tarball and scratch-project checks pin to the worktree the tarball was built from). checks = one {command, exitCode, output} per acceptance criterion: real command, real exit code, output tail. A criterion met by a non-zero exit needs expectExit and a one-line expectReason, only on a single-purpose command, never on a test suite or aggregate runner (run the one failing file alone), never || true. A report task (QA, audit) is done when its checks RAN: FAILs go in result and to release-captain, test -s <report> proves the report exists. A task needing no change still closes with headSha, worktree and its verifying commands. Three refusals ("attempt n of 3") fail the task; a refusal with no evidence object does not count.
Hand-offs. Sections exchange work only as documents: publish yours with org_doc_publish (compact body: counts, SHAs, log paths, one line each; never log bodies), read others with org_doc_read. A document is readable by you only once accepted: otherwise use git and the logs, never wait. No web; only release-captain may ask a human, only in PREFLIGHT.
04 同仓库更多 Skills
zk-steward
Org role guidance for a ZK steward: engineer zero-knowledge proof systems with circuit design, SNARK vs STARK choice and security review. This is zero-knowledge cryptography, not Zettelkasten note-taking.
writing-plans
Use when you have a spec or requirements for a multi-step coding task and must write an implementation plan before touching code. Produces bite-sized TDD tasks with exact file paths, code and commands for an engineer with zero context.
write-spec
Write a feature spec or PRD from a problem statement or feature idea. Use when turning a vague idea or user request into a structured document, scoping a feature with goals and non-goals, defining success metrics and acceptance criteria, or breaking a big ask into a phased spec.
write-query
Write optimized SQL for your dialect with best practices. Use when translating a natural-language data need into SQL, building a multi-CTE query with joins and aggregations, optimizing a query against a large partitioned table, or getting dialect-specific syntax for Snowflake, BigQuery, Postgres, etc.
workflow-automation
Org role guidance for workflow automation: build and maintain CI/CD definitions in GitHub Actions or equivalent: triggers, jobs, matrices, caching. Covers path filters, dependency caching, parallel jobs, timeouts and fail-fast ordering.
workflow-architect
Org role guidance for a workflow architect: map full workflow trees for systems, user journeys and agents, covering every branch and failure. Covers discovering workflows from code, handoff contracts, timeouts, recovery paths and observable state.