Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/agentproto/tsnpx agentmods add skills/agentproto/ts/skillWrote 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/agentproto/ts/skill)<a href="https://agentmods.dev/skills/agentproto/ts/skill"><img src="https://agentmods.dev/badge/skills/agentproto/ts/skill/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/agentproto/ts/skill"><img src="https://agentmods.dev/badge/skills/agentproto/ts/skill.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 | $0.00069 | $0.00943 |
| Opus 5.5 | $0.00028 | $0.00377 |
| Sonnet 5.5 | $0.00014 | $0.00189 |
| Haiku 4.5 | $0.00007 | $0.00094 |
Grade A, and why
session-steward 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 5d 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 — 83 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Session Steward
The steward wraps up idle agent sessions. It classifies them with
session_wrapup_plan, closes the rule-certain ones, and sends the ambiguous
ones through a cheap judge (Jev classifier, fallback: a one-shot LLM agent).
Dry run by default — nothing is closed unless --apply.
Run it
Prefer the CLI (it installs/updates the app and runs the workflow). Three steps, each usable alone, all reading/writing one persisted snapshot:
agentproto steward # 1. classify: rules + Jev, ONE action per session
agentproto steward analyze # 2. LLM reasons for the relevant rows only
agentproto steward act latest # 3. dry run of the planned actions
agentproto steward act latest --apply # …perform them
agentproto steward --apply # one-shot: classify + act
Variants:
agentproto steward classify --llm # classify, then analyze
agentproto steward act latest --rules my-rules.yaml # custom rules (auto-loads ./.agentproto/steward-rules.yaml)
agentproto steward act latest --only mark-failed,needs-input --apply
agentproto steward act latest --session sess_a,sess_b
agentproto steward --legacy --apply --idle 60 # the original single-run steward
Actions: keep | mark-complete | mark-failed | relaunch | needs-input | close-abandoned | archive (relaunch / archive only when named in
--only). act skips a session that changed since the snapshot and never
closes a user-origin session (it flags it). Each command blocks until the
report is printed; --no-wait returns the runId, --json prints the output.
Read the report
The final step output is a markdown table: class (close/stuck/judge), session label + id, idle time, RAM, verdict, confidence, reason, action. Key reading rules:
closerows are safe to auto-close;judgerows are what the judge saw.- Classify rows carry the typed verdict and its probabilities
(
p: active=0.84 blocked=0.08 …); analyze adds the free-text reason. - Verdicts:
done|abandoned|blocked|needs-input|active. Only confidentdone/abandonedclose (sessions stay resumable; transcripts preserved).blocked/needs-inputonly get flagged, never closed. - A judge error never closes anything. If many verdicts come back
activeconfidence 0, the judge lane failed (check Jev key configjev.apiKeyin ~/.agentproto/config.json, envJEV_API_KEYfallback). - RAM freed / still held is summarized at the end.
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.
- 5d ago Changed · +8 lines 909a68202e25
- 13d ago First seen · 75 lines · 69 tokens per session scan A 94bd17aefff1
session-steward is a skill published in the GitHub repository agentproto/ts (10 stars, last pushed yesterday), licensed Apache-2.0. It adds 69 tokens to every session and 943 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-10-03.
Other skills, from other repositories
markedit-tools
Provides tools for managing MarkEdit, a macOS markdown editor.
deferred-task-execution
Delays execution of a task until a specified time or after a duration. Use when the user asks to run something later, in X minutes/hours, at a specific time, schedule a command, or defer work to a future point.
herdr
Control Herdr, a terminal multiplexer for coding agents. Use only when the user explicitly asks to you use or configure Herdr.
cocoplus-off
Deactivate Full Assist Mode — removes all CocoPlus feature flags. Data files (memory, meter history, grove patterns) are preserved. Idempotent.
glyph
Render an answer as ASCII art plus semantic emojis inline with no setup questions: one render per reply, verdict first. Use for any answer with shape: status, inventories, audits, budgets, comparisons, rankings, pipelines, 'what is using X', or any ad-hoc 'show me X visually' ask. Not for definitions, conceptual…
wisdom-forget
Remove a must-keep CocoWisdom entry with rationale. Usage: $wisdom forget --id --rationale " ".