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 skills add ramarahmanda/tsubasa --skill onboardgit clone --depth 1 https://github.com/ramarahmanda/tsubasaWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/ramarahmanda/tsubasa/onboard)<a href="https://agentmods.dev/skills/ramarahmanda/tsubasa/onboard"><img src="https://agentmods.dev/badge/skills/ramarahmanda/tsubasa/onboard/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/ramarahmanda/tsubasa/onboard"><img src="https://agentmods.dev/badge/skills/ramarahmanda/tsubasa/onboard.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00110 | $0.01399 |
| Opus 5 | $0.00055 | $0.00700 |
| Sonnet 5 | $0.00022 | $0.00280 |
| Haiku 4.5 | $0.00011 | $0.00140 |
Grade A, and why
captain-onboard 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 9d ago.
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 — 85 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Captain onboard
Birth a captain in this repo or workspace. Everything through conversation — the user never touches the CLI.
Already a captain here?
If .tsubasa/captain.toml exists, this repo has a captain and init will
refuse. Run tsubasa upgrade instead: it is idempotent and brings a captain
built by an older tsubasa up to date (schema stamp, source graph, memory
tiers), reporting every change. Then stop, and report what changed. Only run
the steps below when there is no captain here.
Steps
- CLI present?
which tsubasa— if missing:uv tool install tsubasa(fallback:uv tool install git+https://github.com/ramarahmanda/tsubasa). Ifuvis missing too, ask before installing it. - One round of questions (single message, not an interview):
- captain name (suggest one from the repo/org name, e.g. captain-)
- role (default: Engineering Director)
- domains (suggest 3-5 from what you see in the codebase)
- Auto-detect sources — scan for:
- git repos:
.githere or in first-level subdirectories (workspace mode) - ADR/design/architecture docs:
docs/adr,docs/decisions,*/docs/*.md(design docs are usually the highest-value source — do not skip them) - postmortems/incidents: dirs or files matching incident/postmortem/outage Present the proposed source list as a table and confirm in one line. NEVER add credential files, .env, or secret stores as sources.
- git repos:
- Scaffold:
tsubasa init <name> --role "<role>" --domains <d1,d2>, then register every confirmed source withtsubasa source add <adapter> <path> [--glob "<glob>"]. NEVER hand-edit captain.toml — the command validates and keeps it parseable. - Persona check. init generates
.tsubasa/persona.mdand includes it fromCLAUDE.mdalongside the hot-memory include: response rules (straight answers, cite or "I don't know", never an id you did not read, the recorded value first, push back on ADR conflicts), the enforced ADR format, and communication rules per Strunk & White's The Elements of Style ("omit needless words"). Read it, verify both include lines are present, do NOT duplicate the rules anywhere, and tell the user in one line that persona.md is generated (tsubasa rewrites it onupgrade) and their own additions go inCLAUDE.mdaround the include. From this step on, ANSWER IN THAT STYLE — the onboarding report itself is the first demonstration of the persona. - First ingest:
tsubasa ingest(deterministic: docs, tags, ADR-marked commits). - Code index — always, not optional.
tsubasa index: deterministic code-only graphify indexes per workspace repo (local AST, no LLM, seconds per repo; non-code files excluded by design). Thentsubasa linkto seed anchors between graph entities and code nodes. This is what lets queries join the why (native graph) to the what-is (code graph). Optional depth, on request only:/graphify ./<repo> --mode deepfor semantic code edges on a repo that warrants it — never a default; docs enter the graph via tsubasa sources, not graphify. - Study phase — this is what makes the captain a veteran. Deterministic
ingest is thin on repos without ADR conventions; now read and distill:
- each detected docs file: extract the decisions/architecture it records
- full git history: run
tsubasa study(headless, chunked — distills EVERY commit into events;--max-chunks Nbounds cost). Thentsubasa resolve(merge duplicate entities) andtsubasa profile(rich profiles for hub entities). This is the 25-year-veteran pass. - architecture scan per repo: README, docker-compose / k8s manifests /
helm values, CI/CD configs, IaC — distill entities for services, envs,
and secret-REFS (name + where it lives, NEVER the value) and relations
(depends_on, deployed_to, built_by, reads_secret).
Do NOT study application source code: structure is answered live from
code (it goes stale silently in a graph); code-level insight enters the
graph organically via recall's learn-on-miss.
Persist each distilled finding as one event with entities/relations:
tsubasa event add --type <decision|note|incident> --title "..." \ --summary "..." --impact <i> --domains <d> --ref <commit|doc>:<id> \ --entity <id>:<type>:<name>:<desc> --relation <src>:<pred>:<tgt>Aim for quality over volume: 15-30 well-connected events beat 200 thin ones.
- Report in captain style (per the step-5 principles): events per source, what the study phase learned (one line per theme), entity count, indexed repos + anchor count, hot-tier size, and 2-3 example questions the user can now ask ("try: why …").
- If this is a git repo, note that
.tsubasa/+CLAUDE.mdare commit-ready (do not commit unasked). In a multi-repo workspace root, note the knowledge lives at the workspace level.
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.
- 9d ago First seen · 85 lines · 110 tokens per session scan A d02d1629118c
captain-onboard is a skill published in the GitHub repository ramarahmanda/tsubasa (8 stars, last pushed 12d ago), licensed MIT. It adds 110 tokens to every session and 1,399 once invoked, about $0.0006 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
decision
Single entry point for all decision operations: capture, search, manage. Routes based on natural language intent.
repo-standard
Use when working in a RepoKit-compliant repo — about to commit or open a PR, adding a learning / ADR, or asking "where does X go" or "what's the convention here". Loads the RepoKit conventions and the pre-commit / pre-PR checklists.
tree-ring-memory
Guides AI agents in using Tree Ring Memory for durable recall, project decisions, user preferences, warnings, future seeds, privacy-safe memory capture, and lifecycle-aware forgetting.
openlore-brainstorm
Transform a feature idea into an annotated story using a Domain Sketch or Constrained Option Tree. Use when asked to brainstorm, explore, or shape a feature before implementation.
mission-brief
Mission-driven SDD orchestrator: take a feature description, structure it into a Mission Brief (goal, constraints, success criteria), generate an ordered step list with prompts that trigger installed SDD skills via model invocation or command-file discovery, and walk those steps to converged implementation. Use when…
levelup-publish
Compile accepted Context Directive Records (CDRs) into team-ai-directives artifacts and create a draft PR. Builds context modules, evals goldensets, and/or skills based on CDR context types.