Communications kit
Templates for the messages a rollout needs: the launch announcement, a pilot invitation, a note to recruit champions, and a series of short weekly tips. Replace everything in angle brackets, cut what does not apply, and write in your own team’s voice. People respond better to a colleague’s plain note than to a polished campaign.
Before sending anything, check that the setup it describes is real on your machines: the managed settings are deployed, the support channel exists, and the first repositories have a SYNTIC.md.
Before you announce
Have answers ready for the four questions everyone asks first:
- What may I use it on? Which repositories, and whether customer data or production systems are off limits.
- What is already protected? The deny rules and other managed settings you deploy, in one or two sentences.
- Who pays, and is there a limit? Whether usage is on a team plan, and whether caps apply.
- Where do I ask for help? One channel, one named person.
Launch announcement
Subject: Syntic Code is available to [team or company]
Hi all,
From today you can use Syntic Code, a coding agent that works in your terminal. It reads the code in your repository, runs commands, and makes changes you review before they land.
Getting started (about ten minutes):
- Follow the setup steps on [internal link].
- Sign in with
syntic auth login, using your work account.- Open a repository you know well and run
syntic. Ask it to explain a part of the code you find confusing.What is already set up for you: [for example: the agent cannot read
.envfiles or private keys, and it asks before running anything that pushes, deploys or deletes.]Use it for: [the repositories and kinds of work in scope]. Please don’t use it for: [anything out of scope].
Questions go to [channel]. [Name] is the contact for this rollout, and each team has a champion who can help: [list].
Thanks, [name]
Short version for a chat channel
Syntic Code is now available for [scope]. Setup takes about ten minutes: [link]. Start by asking it to explain some code you know, then try a small fix. Questions in [channel].
Version from an engineering leader
We are making Syntic Code available because we think it can take routine work off your plate: tracing a bug through unfamiliar code, writing the first draft of tests, keeping a refactor consistent across many files. It does not replace review. Everything it changes goes through the same pull requests and checks as any other change. Try it on real work for the next few weeks and tell us honestly what helps and what doesn’t. [name]
Pilot invitation
Hi [name],
We are running a [two/four]-week pilot of Syntic Code with a small group before offering it more widely, and I’d like you in it.
What it involves:
- Use it on your normal work where it seems useful. No quota.
- Post in [pilot channel] when something works well, or when it gets in your way.
- A 15-minute check-in each [day].
- A two-question survey at the end.
What you get: early access, and a real say in how we set it up for everyone else.
Reply by [date] if you’re in.
Recruiting champions
Hi [name],
You’ve been getting good results with Syntic Code in the pilot. Would you be the go-to person for it on [team] as we roll it out?
In practice that means answering the occasional question, sharing something useful you find every week or two, and telling me when something about the setup gets in people’s way. Probably an hour a week at first, less later.
There’s a short guide here: [link to your copy of the champion kit]. Happy to talk it through.
Weekly tips
One tip a week, kept short, works better than a long guide nobody reads. Each tip below names a real command or setting and says why it matters.
| Week | Tip |
|---|---|
| 1 | Start with a question, not a task. Ask “how does <module> decide <thing>?” before asking for changes. You learn the code and see how the agent reasons. |
| 2 | Give it your project’s rules. Run /init in a repository to draft a SYNTIC.md, then edit it. Every session reads it. See Memory. |
| 3 | Say how to check the work. “Run <test command> when you’re done” turns a guess into a verified change. |
| 4 | Plan before big changes. Switch to plan mode with Shift+Tab and agree the approach before any file is touched. See Permission modes. |
| 5 | Stop allowing the same command over and over. Allow it once for the project with a rule in .syntic/settings.json, or from /permissions. See Permission rules. |
| 6 | Keep sessions focused. /clear between unrelated tasks; /compact when a long session slows down. Less context, faster and cheaper replies. |
| 7 | Undo with confidence. Changes can be rewound to an earlier point. See Checkpointing. |
| 8 | Connect a tool you use every day. MCP servers let the agent read your issue tracker or docs. See MCP servers. |
| 9 | Let it review before you do. /review looks over your current changes or a pull request and points out problems. |
| 10 | Script the boring parts. syntic -p "<task>" runs once and exits, which fits CI jobs and shell scripts. See Headless mode. |
| 11 | Automate a habit. A hook can format files after every edit. See Hooks. |
| 12 | Share what works. Put a good prompt or a skill in the repository so the whole team can use it. See Skills. |
Answers to common questions
Does it change code without asking? Not by default. It asks before editing files and before running commands that are not on an allow list. Your team decides how much to allow.
Can it see secrets? Files matched by deny rules cannot be read. [Describe the rules you deploy.] Keep secrets out of prompts as you would in any chat tool.
Where does my code go? The files and command output the agent reads are sent to the Syntic platform to run the model. Details are on Data usage.
What if it’s wrong? Treat its changes like a colleague’s draft: read the diff, run the tests, and review the pull request as usual.
Does it work on Windows? Yes. A few features differ by system; see Feature availability.
How do I see what I’ve used?
/usage shows your plan’s usage; /cost shows the current session’s estimated cost.
Prompt starters to share
Short, concrete prompts give people a quick first success:
- “Explain what happens when a request reaches
<endpoint>, file by file.” - “Find where
<setting>is read and list every place that depends on it.” - “Write tests for
<function>covering the edge cases you can find. Run them.” - “This test fails:
<paste>. Find the cause and propose the smallest fix.” - “Review my staged changes for bugs and missing tests.”
- “Update
<old API>to<new API>across the repository. Show me the plan first.”