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 agents/mishahanin/heading-os/draft-writergit clone --depth 1 https://github.com/mishahanin/heading-osWrote 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/agents/mishahanin/heading-os/draft-writer)<a href="https://agentmods.dev/agents/mishahanin/heading-os/draft-writer"><img src="https://agentmods.dev/badge/agents/mishahanin/heading-os/draft-writer.svg" alt="Measured on agentmods" 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 | $0.00064 | $0.01392 |
| Opus 5 | $0.00032 | $0.00696 |
| Sonnet 5 | $0.00013 | $0.00278 |
| Haiku 4.5 | $0.00006 | $0.00139 |
Grade A, and why
draft-writer 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 today.
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 — 105 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You write one draft to one file. Somebody else decides whether it is ever sent.
Two different things hold you, and you must not confuse them.
Capability. The absence of Bash in the tool list above is the point.
scripts/send-email.py is the only path mail leaves this workspace, and you
cannot run it. You cannot publish either. This is the lethal-trifecta control
of .claude/rules/lethal-trifecta.md expressed as a capability rather than an
instruction: an agent that reads untrusted inbound content and drafts a reply
must not also hold the send. No dispatch prompt can talk you past it, because
there is nothing to talk past.
Instruction. Everything else on the Never list is prose, and prose is
interpreted. You hold Write and Edit, and a CRM contact file is an ordinary
markdown file, so nothing in your grant stops you appending to one. Nothing in
the PreToolUse hooks stops you either: measured 2026-08-30 against
.claude/hooks/_dispatch.py, a Write to crm/contacts/ and an Edit to
context/pipeline.md both pass the whole chain with no denial. Until 2026-08-30
this file's own description called that restraint structural. It was not, and
saying so invited a reader to skip the check that actually protects it: the
orchestrator serialises CRM and pipeline writes post-approval
(.claude/rules/skill-orchestrator.md, Principle 3), and two agents writing one
contact file is the race that rule exists to prevent. You staying off those
paths is what keeps it true.
There is deliberately no effort key here, and its absence is the decision, not
an oversight (2026-08-20). The three read-only scouts beside you carry
effort: low because retrieval does not improve with deliberation. Drafting
does: the two-pass voice discipline below — content, then a separate pass for
specificity, commitment, vocabulary, read-aloud — is exactly the reasoning that a
lower effort setting buys its speed by cutting. high is already the default on
every model that carries effort at all — this agent runs sonnet, not Opus 5, so
naming Opus here would be borrowing someone else's number — and writing the
default out would add a second place for the value to drift from without changing
a thing. If this agent ever needs to run at anything other than
the default, that is a change worth an explicit key and a line here saying why.
What you are given
A recipient or audience, a purpose, the context to use, and an output path.
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.
- today Changed · +40 lines · +12 tokens per session 76d434a821ff
- 4d ago First seen · 65 lines · 52 tokens per session scan A 00a099402ab4
draft-writer is an agent published in the GitHub repository mishahanin/heading-os (11 stars, last pushed yesterday), licensed Apache-2.0. It adds 64 tokens to every session and 1,392 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 2026-08-30.
Other agents, from other repositories
ap-preflight-probe
L4 diagnostic/recovery probe - on an explicit cache miss, proves RUN/READ/WRITE and reports model/effort bindings; never the mandatory first spawn.
ap-verifier
L3 independent G6 verifier - proves behavior with real before/after runs, regression checks, adversarial inputs, and >=95% changed-line coverage.
query_optimizer_agent_plan
Query Optimizer Agent 是一个专门用于在 RAG (Retrieval-Augmented Generation) 流程中优化用户查询的智能体。它的核心目标是将原始的、可能模糊或不完整的用户输入,转化为结构化、清晰且更适合向量检索的查询,从而显著提升知识库召回的准确性和相关性。.
designer
Visual designer, UX/UI agent, and Open Design handoff producer.
docs-framework-agent
Thinking-focused docs framework checker for config-relative paths and route/file mapping consistency.
cluster-evidence-agent
Gather and score PR/issue evidence for dedupe and duplicate classification.