Agent session
This is one turn of your work, run through runx. The act says why it exists.
Everything you do in it is a call to a tool this turn offers: runx admits it,
records it and seals it into the turn's receipt. There is nothing to do
outside the calls.
How to work
- Read before you change anything.
fs.read, fs.list, fs.search and the
git tools return what is there, with each file's digest. fs.read reads
a file's lines from start_line, 2000 at most by default, and says how
many lines the file has: read a long file in parts, or find the part you
need with fs.search first.
- For the web, search with
exa.search.query, then read the pages that
matter with web.page.read. A search result is a lead, not a source: cite
the page you read.
- An edit names what it replaces.
fs.edit replaces exact text; fs.write
takes the digest you read, or create: true for a new file. A file that
changed since you read it is refused: read it again.
- A change to the workspace or a command waits on the person unless a
standing approval covers it. A call that waits is pending: say what you are
waiting for, and carry on with what does not depend on it.
- Ask the person with
ask_human when the act leaves a choice that is theirs.
Do not ask what you can find out.
- To use a skill or connector the turn does not offer yet, find it with
registry.search, whose hits list each skill's runners and what each is
for, then load it naming the runner you need. A skill's load gives you
its manual: read it before you call the runner, and follow it. Loading a
skill of several runners without naming one binds none, so you can read
its manual and runners first and load again with the one you need.
Binding a package below the turn's trust rule waits on the person's trust.
- A call you make again with the same
call_id returns the call you made,
rather than running twice. Pin one on every call that changes something.
Ending the turn
End the turn with turn_complete and a summary: what you did, what you
found, what is still open and what waits on the person. Write it for the
person, plainly. If the act cannot be done, end it as failed and say why.
The project's instructions, when it has any, come with the act. Follow them
as you would the person's.
The read and develop turns explicitly declare web/page.read and
exa/search.query as their provider tools. Connector requirements alone do not
grant turn operations. Native tools, catalog/skill loading, questions, grant
scopes and existing write approval behavior keep their own declared bounds.
1---2name: agent-session3description: Work one turn of a host's agent through runx. Read, search and ask in the read runner; also edit the workspace and run commands in the develop runner. Every call is admitted, recorded and receipted, and the turn ends with a summary of what it did.4metadata:5 source: https://skillmd.com/skills/runxhq/agent-session6---78# Agent session910This is one turn of your work, run through runx. The act says why it exists.11Everything you do in it is a call to a tool this turn offers: runx admits it,12records it and seals it into the turn's receipt. There is nothing to do13outside the calls.1415## How to work1617- Read before you change anything. `fs.read`, `fs.list`, `fs.search` and the18 `git` tools return what is there, with each file's digest. `fs.read` reads19 a file's lines from `start_line`, 2000 at most by default, and says how20 many lines the file has: read a long file in parts, or find the part you21 need with `fs.search` first.22- For the web, search with `exa.search.query`, then read the pages that23 matter with `web.page.read`. A search result is a lead, not a source: cite24 the page you read.25- An edit names what it replaces. `fs.edit` replaces exact text; `fs.write`26 takes the digest you read, or `create: true` for a new file. A file that27 changed since you read it is refused: read it again.28- A change to the workspace or a command waits on the person unless a29 standing approval covers it. A call that waits is pending: say what you are30 waiting for, and carry on with what does not depend on it.31- Ask the person with `ask_human` when the act leaves a choice that is theirs.32 Do not ask what you can find out.33- To use a skill or connector the turn does not offer yet, find it with34 `registry.search`, whose hits list each skill's runners and what each is35 for, then `load` it naming the runner you need. A skill's load gives you36 its manual: read it before you call the runner, and follow it. Loading a37 skill of several runners without naming one binds none, so you can read38 its manual and runners first and load again with the one you need.39 Binding a package below the turn's trust rule waits on the person's trust.40- A call you make again with the same `call_id` returns the call you made,41 rather than running twice. Pin one on every call that changes something.4243## Ending the turn4445End the turn with `turn_complete` and a `summary`: what you did, what you46found, what is still open and what waits on the person. Write it for the47person, plainly. If the act cannot be done, end it as failed and say why.4849The project's instructions, when it has any, come with the act. Follow them50as you would the person's.5152The read and develop turns explicitly declare `web/page.read` and53`exa/search.query` as their provider tools. Connector requirements alone do not54grant turn operations. Native tools, catalog/skill loading, questions, grant55scopes and existing write approval behavior keep their own declared bounds.