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 zookanalytics/gc-toolkit --skill converse-holdgit clone --depth 1 https://github.com/zookanalytics/gc-toolkitWrote 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/zookanalytics/gc-toolkit/converse-hold)<a href="https://agentmods.dev/skills/zookanalytics/gc-toolkit/converse-hold"><img src="https://agentmods.dev/badge/skills/zookanalytics/gc-toolkit/converse-hold/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/zookanalytics/gc-toolkit/converse-hold"><img src="https://agentmods.dev/badge/skills/zookanalytics/gc-toolkit/converse-hold.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 | $0.00017 | $0.01596 |
| Opus 5.5 | $0.00007 | $0.00638 |
| Sonnet 5.5 | $0.00003 | $0.00319 |
| Haiku 4.5 | $0.00002 | $0.00160 |
Grade A, and why
converse-hold 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 3d 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 — 126 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Step 5 — Hold
Stamp what you are waiting for, then post your framing. $CONV, $VISIT,
and $SUBJECT are resolved in step 1's claim block.
converse-hold.sh takes the one decision or input needed as its argument
and $VISIT / $SUBJECT in its environment. It exits non-zero when the
hold did not fully land, and then you must NOT frame:
if VISIT="$VISIT" SUBJECT="$SUBJECT" \
"$CONV/converse-hold.sh" "<the one decision or input needed, ≤130 chars>"; then
: # the hold is real and stamped — post the framing below
else
# NOT a hold yet: nothing re-asks the subject. Do NOT post the framing.
# Raise the failure in the thread and do not describe the subject as held.
exit 1
fi
A hold IS a demand. The operator owes an answer before the
conversation can conclude, so the wait is a bead with a blocks edge,
not a comment, and converse-hold.sh files it against the right bead for
you. A conversation about a PR anchor gates the VISIT: the conversation
cannot conclude until the operator answers, and the subject anchor keeps
moving — a conversation does not freeze its subject's merge. Only a
pre-PR (unanchored) subject takes the demand on itself, because its held
marker needs that edge. A ruling files unassigned and routes to the
operator's partition; pass --assignee <who> to the writer only when the
demand is work a named person must perform, and that one is theirs to
close, never yours. A resumed hold refreshes the demand this sitting
filed rather than filing a second. The visit's escalation_key scopes the
demand, so a sibling sitting with its own key on a shared subject keeps its
own.
Stamp BEFORE you wait, not after. The hold IS a demand: until the
demand is filed nothing re-asks the question, so it lands before you hand
control to the operator. Write the takeaway to state the decision needed
when read cold off the board. The same step sets gc.hold_demand on this
visit, the trace step 1's action=hold arm reads to tell a real hold
from a claim that died before step 2.
To pause the merge, pass --hold-merge — by default a conversation does
not. The hold above leaves a PR free to land while you talk, which is the
shepherd case a conversation opened to help a stuck PR wants. When the sitting
instead decides the merge must wait on the operator — the PR should not land
until this is settled — add the flag to the same call:
if VISIT="$VISIT" SUBJECT="$SUBJECT" \
"$CONV/converse-hold.sh" --hold-merge "<the one decision or input needed, ≤130 chars>"; then
: # the hold is real and the merge is held — post the framing below
else
# the conversation hold or the merge hold did not land — do NOT frame.
exit 1
fi
It files a second demand on the anchor — the blocks edge the merge sweep
already honors — and fails the hold closed if that demand does not land, so a
framing never claims a merge hold it did not take. Omit the flag and the PR
keeps moving. Either way the step-7 sign-off discharges whichever demands you
filed.
The takeaway is the sentence; held is the state. Where $SUBJECT
already carries an anchor state the held transition is skipped, and
refused if attempted: merge.sh, gate-ensure.sh and pr-facts.sh
enumerate anchors by that state, and held drops it from all three. It
is the pre-PR hold; the explicit merge hold above is an edge on the
anchor, not this state.
A framing that asks for no decision still files one. What the operator owes then is the close-out itself, and the demand is what brings the subject back if the thread is lost before they take it. The gate is not about there being a question; it is about the subject not moving until a person acts.
One sentence, ≤130 characters. The takeaway reads
holding — <your sentence>, the board's NEEDS cell, and the prefix spends
part of the headline cap. converse-hold.sh refuses a longer sentence
before it writes anything and says how far to cut it. What will not fit
goes in the notes.
Never park a live conversation: the writer's --release clears the
assignee and route, and the only place it belongs is a stand-down
ruling, written as
gc-helm.sh takeaway <anchor> "<ruling>" --release --no-wait. It parks
the anchor AND cancels the run beneath it through gascity, and --no-wait
records that the ruling ended the wait rather than moving it.
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.
- 3d ago Changed 092b1196f9b2
- 6d ago Changed · +3 lines 41a408f00c8a
- 9d ago Changed · +2 lines 8c3acba61663
- 10d ago Changed · +1 lines cac622229fa9
- 12d ago Changed · +1 lines 872e65fbabc2
- 13d ago Changed · +26 lines 2a3b56d1f5c8
- 20d ago Changed · +2 lines cccc34043744
- 26d ago First seen · 91 lines · 17 tokens per session scan A d1e6206b881c
converse-hold is a skill published in the GitHub repository zookanalytics/gc-toolkit (5 stars, last pushed today), licensed MIT. It adds 17 tokens to every session and 1,596 once invoked, about $0.0001 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 2026-09-22.
Other skills, from other repositories
oma-recap
Summarize AI conversation histories for a specified date or period. Use for daily work recaps and cross-tool activity summaries.
tmux
Manage long-running shell sessions with tmux — start a detached session, run a long task, reattach later, capture output. Use when a task takes longer than a single tool call (build, test, log tail).
himalaya
CLI to manage emails via IMAP/SMTP. Use himalaya to list, read, write, reply, forward, search, and organize emails from the terminal. Supports multiple accounts and message composition with MML (MIME Meta Language).
cron
A scheduler for tasks that should run automatically at a future time or repeat on a schedule. It uses cron expressions, a text format for describing times such as every day at 9:00.
apple-notes
Manage Apple Notes via memo CLI: create, search, edit.
template-engine
Load, validate, and populate templates consistently across frameworks.