01 / Purpose
A writing system your team owns.
Chalkline helps your team write down its voice, terminology, and channel rules so agents can consult them before drafting. You review the rules and keep them in editable Markdown files.
This site has no live model or uploads. The receipt example illustrates the traceability contract; it does not verify a draft.
02 / Drift
Keep writing guidance out of scattered prompts
When writing guidance is copied into separate prompts for support replies, product messages, and release notes, those copies can drift apart. Chalkline keeps the guidance in one set of references and instructs agents to consult them before drafting, flag missing requirements, and ask about unresolved conflicts. Product facts, deadlines, and promises still need their own sources; writing rules cannot establish them.
03 / Setup
Set up your references
Open your own copy of Chalkline in an agent that reads repository instructions and ask:
“Set up my writing system.”
Answer a few questions.
The agent asks about your team’s voice and terms, one question at a time. Share only material you have permission to use.
Review the proposed rules.
Review each proposed rule, including any extracted from your examples, before approving it for your team’s use.
Compare the rewrites.
Compare rewrites of a sample with and without the references. Decide whether you approve the draft and, separately, whether it should guide similar work.
Start privately unless all inputs and outputs are safe to publish. Review changes before choosing whether to commit.
04 / Files
What the files contain
Setup produces two to four reference files, depending on the guidance your team needs.
- voice.md
- How you sound, with examples.
- terminology.md
- Preferred words and non-negotiable terms.
- vocabulary.md optional
- Maps everyday requests to your team’s terms so agents can understand them without correcting how people speak.
- channels.md if needed
- Guidance specific to a channel.
AGENTS.md tells agents how to consult the system. Approved before-and-after examples show how the guidance applies to a particular audience, channel, and task.
05 / Use
Inspect a draft and its sources
Choose a fictional Harbor task to see its facts, writing rules, draft, and receipt. The examples are prewritten, not generated or verified by AI. Removing a required source shows how a blocked task should be reported; nothing is sent to a server.
Choose a task
Supplied facts
Applicable rules R1–R2
Candidate draft
The required terminology source is missing. Restore it before drafting.
Example copy; not approved for use.
Usage receipt
Self-report / separate from copyThis example receipt shows what an agent would report. It does not record an actual review or approval.
0 blocked · 1 limited · 2 supported · 3 corroborated. Use the lowest applicable score, never an average. The score describes the evidence available for this task; it does not establish correctness, compliance, or permission to publish.
06 / Improve
Revise guidance through review
If a draft reveals missing or unclear guidance, propose a reference change for the responsible person to review. An accepted edit does not automatically become a rule for everyone. Test changes on new, comparable work without showing the agent the expected rewrite; the sample used during setup cannot establish whether the guidance helps on other tasks.
Get started
Use the setup protocol and fictional examples in the repository to build your own references, with reviewers chosen by your team.
Explore Chalkline on GitHub