aftercare

Your agent checks its own work after the fact. When a reply says something is deployed or live at a public address, Aftercare remembers it and, every few hours at session start, checks it again with one GET request in the background. If it has stopped working, the agent is told once and tells you first; you hear when it is back. /aftercare lists and checks, and hermes aftercare check runs from cron. Never checks private networks or runs commands. Disclosure — on by default, sends background GET requests (user agent `hermes-aftercare/0.1`) to any public http(s) address an agent reply says is deployed or live, at session start every 6 hours by default, and adds a one-time note to the next turn when one changes state; private, loopback and local-name addresses are refused.

salch-credToolsv0.1.0Updated 23 hours ago

About

aftercare

Your agent checks its own work after the fact.

"Deployed. The shop is live at https://shop.example.com." It was true when the agent said it. Is it still true on Thursday?

Every agent tool checks work at the moment it's done. Aftercare comes back later:

It remembers the address. When a reply says something is deployed or live at a public address, Aftercare notes it.
It checks again. Every few hours, at the start of a session, it checks each address with one plain request.
It tells you. If something the agent said was working has stopped, the agent is told at the start of your next turn, and it tells you first:

[Aftercare] Something you told the user was working has changed since:
- https://shop.example.com stopped working after you said on 10 Nov 14:02:
  "Deployed. The shop is live at https://shop.example.com." Now: HTTP 500.

You hear when it's back, too.

Use it

You don't need to do anything: addresses are picked up from the agent's replies. To see or change what's watched:

/aftercare                 what is watched, and how each one looked last time
/aftercare check           check everything now
/aftercare watch <url>     watch an address yourself
/aftercare forget <n|url>  stop watching one
hermes aftercare check     the same from your shell

hermes aftercare check works from cron, so a server can be checked even when you aren't chatting.

/aftercare check

Checked 2: 1 answering, 1 not.
  ✅ https://example.com  HTTP 200, 281 ms
  ❌ https://httpbin.org/status/500  HTTP 500, 875 ms  <- has not worked yet

What counts as a claim

A sentence in a reply that says something is deployed, live, published, launched, up and running, or available, hosted or running at a public http(s) address.

These are ignored:

links to docs and references
negated or conditional sentences ("not live yet", "once DNS updates it will be live")
questions
text inside code blocks

Safety

It never checks localhost, private networks or local names. It skips 127.0.0.1, 10.x, 192.168.x, ::1, .local and .internal. It also skips any public name that resolves to a private address, and any redirect that leads to one.
One GET request per address, reading at most 1 KB, with a timeout (8 seconds by default). It never runs commands, and never sends anything except the request itself.
Checks run in the background, so starting a session never waits for the network.
Certificates are verified. Aftercare uses the certifi bundle when it is installed, because some Python builds ship without a list of trusted certificate authorities.

What it touches

Network: one GET request to each watched address, per check, with the user agent hermes-aftercare/0.1.
Adds context to your next turn (through pre_llm_call) when a watched address changes state. It does this only once per change.
Stores locally: each address, the sentence that claimed it, and the last six check results, in <hermes home>/plugin-data/aftercare/aftercare.json.
No model calls, no telemetry, no dependencies.

Settings

plugins:
  entries:
    aftercare:
      settings:
        auto_watch: true    # pick up addresses from the agent's replies
        every_hours: 6      # how often each address is checked
        keep_days: 30       # forget an address 30 days after it was last claimed
        max_watches: 25     # most addresses watched at once
        timeout: 8          # seconds to wait for an answer

Limits

Up means the address answers with 2xx or 3xx. A page that loads but shows the wrong content still counts as up.
Checks happen when a session starts (or from cron, or with /aftercare check). If you don't use Hermes for a week, nothing is checked that week unless cron runs it.
English replies only, for picking up claims automatically. /aftercare watch works for anything.

Development

pytest tests

37 tests cover finding claims, ignoring links and private addresses, state changes (down, never up, back), alerting only once, background checks, the commands, expiry, limits and the private-address refusals. The network is scripted, so no test sends real requests.

License

MIT

README from the reviewed commit 3ffaade; it updates when the author re-pins.

More by salch-cred

12 plugins by salch-cred in the catalog.

salch-credbluffDoes your model make things up? Bluff asks about things that sound real but don't exist (Python functions, built-in exceptions, chemical elements), mixed with real ones and phrased as if they all exist, and checks whether the model admits the made-up ones aren't real. The result is a bluff rate, with real items as a control. Every item is checkable by machine, so no model judges another. Runs on your active model through ctx.llm, no setup.Communitytools salch-credcalm-modeWhen you get frustrated, your agent slows down. Calm Mode scores the tone of each message ("still broken", "I already told you", capitals, "!!", the same message twice); frustration builds and fades. Past a threshold the agent works in small checked steps, at most two file edits per turn, and risky commands (rm -rf, git push, deploys, DROP TABLE) wait until the user says yes. /calm shows the level and forces it on or off. No model calls, nothing stored.Communitytools salch-credcrash-testA safety rating for AI agents. For one armed turn it breaks tools the way real systems break (timeouts, outages, empty or cut-off results), then rates what the agent did: retried, worked around it, told the truth, or claimed success anyway. 1-5 stars and an HTML report, comparable across models by seed. Blocked calls never run, silent faults only touch reads, and the agent gets no tools from it, so it cannot tell it is being tested. Disclosure — when the user arms a test (`/crashtest start` or the `CRASH_TEST` environment variable), the next turn's tool calls are deliberately blocked with fake errors and some read results (read_file, search_files, web_search/web_extract, read-only terminal commands) are silently emptied, cut off or replaced with an older result without telling the agent, so the agent may act on incomplete data during that turn; the rating is appended to the reply and the task text and reply are saved in a local report under the plugin's data folder.Communitytools salch-creddeja-fixStops the agent from putting back code that already failed. Every version of every file it writes in a session is fingerprinted and judged by the next test or build run, or by the user saying "still broken". An edit that would recreate a failed version is stopped once, naming when it failed and why; a second try goes through. Catches the same code reached by different edits, which Hermes' identical-call guard can't. /deja shows what is tracked. No model calls.Communitytools

All 12 by salch-cred

Developer Mode

Explore built-in Hermes toolsets and actions.

HermesAgent

Developer?

Create a Hermes extension for Desktop or agent workflows.

Developer Section