Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/romerma/mstack/orchestratenpx skills add romerma/mstack --skill orchestrategit clone --depth 1 https://github.com/romerma/mstackWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00049 | $0.00648 |
| Opus 5 | $0.00024 | $0.00324 |
| Sonnet 5 | $0.00010 | $0.00130 |
| Haiku 4.5 | $0.00005 | $0.00065 |
Grade A, and why
orchestrate scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 57 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Orchestrate
Check the threshold first. Work one agent could finish inside a single session is not a program. This machinery costs real throughput: in one measured head-to-head, the equivalent ceremony turned a twelve-unit job into one landed unit while a plain agent landed all twelve. Below that line, use the feature route.
Choosing this is a decision worth recording, including why the simpler route was rejected:
mstack decide --phase frame --decision "run as a program" --why "..." --evidence "..." --result open
The full procedure is in the orchestrate playbook. Three things carry most of the weight.
Isolation is what makes parallelism legal
mstack worktree new <slug> per concurrent unit. Each worktree carries its own .mstack/,
which is what makes "one active item" mean one active item here rather than one across the
whole machine. Record the base SHA in that worktree's current.md: "branched from main" is not
a checkpoint, because main moved.
The brief is the product
A vague brief fails quietly, because a worker cannot ask you a question. Nine fields: goal, scope, context, acceptance, verify, timebox, forbidden, report, standing constraints. Missing fields are a reason not to launch it yet.
Size the brief to the unit. A four-kilobyte scaffold around a two-line edit costs more to write and obey than the edit.
Workers cannot see each other. Anything a unit depends on gets pasted into its context in full, not referenced as "see the other agent's report".
Landing is continuous, never a final phase
Integration starts with the first verified unit. Walk up from the lowest unmerged change and stop at the first without a passing verdict: a verified change sitting above an unverified one is not landable, because merging it pulls the gap in underneath it.
Concurrent subagents cap at twenty per session, and CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS moves
that number. Past the cap, spawning fails — it does not queue — so mstack fanout plan
refuses first, against the session's configured limit. It bounds one fan-out, not a session: two
plans of fifteen both pass, and keeping the session total under the cap is yours to do.
Fan out to what the work needs, not to what the limit allows, and say what you dropped if you
bound the fan-out: silent truncation reads as full coverage.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 57 lines · 49 tokens per session scan A 023c871b4f4c
orchestrate is a skill published in the GitHub repository romerma/mstack (1 stars, last pushed 8d ago), licensed MIT. It adds 49 tokens to every session and 648 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
happiness-skill
当用户问「怎么才能更幸福/为什么得到了还不满足/怎么减少焦虑」时调用。 核心理念: 幸福是缺憾感清空的默认状态, 是可训练的技能; 欲望是与自己的契约(得到前不快乐), 同时只留一个重大欲望; 活在当下。 不适用于: 临床抑郁等需要专业治疗的场景(本书方法不能替代医疗)。 Triggers: 幸福/不快乐/欲望/焦虑/知足/活在当下/happiness/desire/anxiety.
gsd-audit-milestone
Audit milestone completion against original intent before archiving.
spec-kitty-charter-doctrine
Run charter interview, generation, context, and sync workflows for project governance in Spec Kitty 3.x. Access doctrine artifacts programmatically via DoctrineService. Resolve agent profiles. Load action-scoped governance context iteratively, not all at once. Triggers: "interview for charter", "generate charter"…
spec-kitty-spdd-reasons
Drive REASONS Canvas authoring and review for Spec Kitty missions that opted in to Structured-Prompt-Driven Development (SPDD) via charter selection. Triggers: "use SPDD", "use REASONS", "generate a REASONS canvas", "apply structured prompt driven development", "make this mission SPDD". Does NOT handle: enforcing SPDD…
dependency-upgrade
Plan, batch, and verify dependency upgrades safely. Triages outdated packages into risk tiers, upgrades in order (dev/minor/patch first, runtime majors last), verifies each batch, and produces an auditable commit sequence. Use when asked to "upgrade deps", "bump packages", "update nodemodules", "fix vulnerabilities"…
forensics
Post-mortem a failed GSD auto-mode run. Traces symptom to root cause via .gsd/ activity, journal, metrics, and lock artifacts, producing a filing-ready bug report with file:line refs and a fix suggestion. Use when asked to "forensics", "post-mortem", "why did auto-mode fail", "trace the stuck loop", "debug the crash"…