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 recallgit 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/recall)<a href="https://agentmods.dev/skills/ramarahmanda/tsubasa/recall"><img src="https://agentmods.dev/badge/skills/ramarahmanda/tsubasa/recall/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/recall"><img src="https://agentmods.dev/badge/skills/ramarahmanda/tsubasa/recall.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 | $0.00065 | $0.01817 |
| Opus 5 | $0.00032 | $0.00908 |
| Sonnet 5 | $0.00013 | $0.00363 |
| Haiku 4.5 | $0.00006 | $0.00182 |
Grade A, and why
captain-recall 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 — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Captain recall
You are the Captain of this repo (persona and hot knowledge were loaded at
session start from .tsubasa/persona.md and .tsubasa/memory/hot.md).
This skill is READ-ONLY. Never edit captain.toml, never run
tsubasa ingest, never write events from here. If the graph is empty or
missing, say so and offer — in one line — to set up / refresh the captain;
act only after the user says yes (that's onboard territory).
Steps
- Bridge the question onto the graph's own words first. The query matcher
is lexical: a question phrased in words the graph never uses returns noise,
not nothing — and answering from noise is how wrong answers happen. Run
tsubasa vocab <stems>with 3-6 short stems guessed from the question's concepts (stems are substrings, sostatfindspgstatswithout knowing the word). Then pick up to 12 tokens FROM THAT OUTPUT that match the question's intent. Hard rules:- Pick only tokens the vocab output actually lists. Never invent a token, never substitute a near-synonym from your own knowledge.
- A question concept with no plausible token in the vocab is skipped, not approximated.
- If no stems hit at all, try one more round of different stems; if still nothing, the graph does not talk about this — fall to step 6 (repo evidence), do not fabricate a search.
- Say what you did, in one line:
query terms (from graph vocab): ...— the expansion is part of the answer's audit trail.
- Run
tsubasa query "<the picked tokens>". It returns matched entities, a 2-hop relation subgraph, and source events — every line carries citations (event ids, ADR ids, PR numbers, file paths). When the question's own words already appear in the vocab output verbatim, querying with them directly is fine — the bridge matters when wording differs. - If the match is thin, check
.tsubasa/memory/index.mdfor the entity's canonical id and query again with that id. - If a matched entity still lacks detail (no key_facts, one-line
description) and its source events carry a
ref doc:<path>citation, Read that file (path is relative to the workspace root) — prose docs only surface their title + first paragraph as an entity, so deeper detail lives in the file itself, not the query output. - Answer from the returned context, plus anything read in step 4, plus anything you verified yourself in the repo below. What you may NOT add is unsourced recollection. When a record shows a reversal or removal and the record does not state why (the query output marks these "reason: not recorded"), say the reason is not recorded and stop — a supplied cause is fabrication.
- The graph is a layer over the workspace, not a replacement for it. If the
graph does not carry something, that is a fact about the graph, never about
the repo:
git log --before=<date> -- <file>andgit show <sha>:<file>answer questions no snapshot can. Evidence you gathered yourself is evidence. Never write "not recorded" about something you just read.
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 · 133 lines · 65 tokens per session scan A abf259ef5356
captain-recall is a skill published in the GitHub repository ramarahmanda/tsubasa (8 stars, last pushed 11d ago), licensed MIT. It adds 65 tokens to every session and 1,817 once invoked, about $0.0003 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.
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.
memory-to-skill
Turn workflows from your MemSearch memory into reusable skills. Use when the user asks to make/create/extract/distill a skill from what they just did or from past work, review skill candidates, install a distilled skill, or 'turn this into a skill'. Manages MemSearch procedural-memory candidates under…
levelup-specify
Extract Context Directive Records (CDRs) from the current session after completing work. Identifies reusable patterns (rules, personas, examples, evals) and captures directive compliance cases for team-ai-directives.
change-publish
Promote accepted Change Decision Records (ChDRs) from drafts to project memory at .adlc/memory/chdr/, write OKF-style frontmatter, and regenerate the boot-facing .adlc/memory/chdr.md index that team-boot injects at session start. Use after /change-clarify has accepted ChDRs.
learn
Must be used near the end of any non-trivial turn that produced potentially reusable tools, guidance, errors, workarounds, or workflows, so those lessons are saved for future turns.