Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 13 additions & 1 deletion AGENTS.md
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,7 @@ and its resources. The source layout stays under `app/`.
`feed`: the live feed any Harbor writes, from its own trial hooks),
`importers/` (`atif` ATIF, `otel` OpenTelemetry, and the `cli`), `judge/` (`questions`, `state`, `client`, `store`, `batch`, `reconcile`, `lanes`),
`standard/` (the Nullius standard, `docs/standard.md`: `ports`, `names` (capability vocabulary and legacy aliases), `definitions` (with an agent's or capability's versions), `match`,
`jsonl`, `acp`, `cancellation` (interruptible protocol waits), `relay`, `ledger` (a run's spending cap), `pins` (image digests checked against registry tags), `install`, `run`, `container` (the container backend), `setup_container` (an Original setup's driver and its Linux dependencies in Docker), `forward` (a licence server's one door from it, or a benchmark's MCP server's), `web_services` (HTTP fixtures a driver's setup declares, each behind a unit-local Docker door), `mcp_machine` and `mcp_bridge` (a benchmark's MCP server run on the machine, in its driver's environment, and the agent's bridge to it), `network_requirements` (an environment's declared public web access: the schema and the connectivity check), `hostreq` (what a benchmark's options need of the machine its driver runs on: the check and the install) and `egress` (an own image's check's proxy to public addresses), `job`, `conform`, `fidelity` (ACP against the agent's own CLI), `prove` (a benchmark's controls and proofs), `shard` and `merge` (one run over many machines), `recipe` (the builder's standard recipe), `howto` (the How to add steps) and the `cli`; its
`jsonl`, `acp`, `cancellation` (interruptible protocol waits), `relay`, `ledger` (a run's spending cap), `pins` (image digests checked against registry tags), `install`, `run`, `container` (the container backend), `boat` (the boat backend: each unit's agent in a boat.dev VM), `setup_container` (an Original setup's driver and its Linux dependencies in Docker), `forward` (a licence server's one door from it, or a benchmark's MCP server's), `web_services` (HTTP fixtures a driver's setup declares, each behind a unit-local Docker door), `mcp_machine` and `mcp_bridge` (a benchmark's MCP server run on the machine, in its driver's environment, and the agent's bridge to it), `network_requirements` (an environment's declared public web access: the schema and the connectivity check), `hostreq` (what a benchmark's options need of the machine its driver runs on: the check and the install) and `egress` (an own image's check's proxy to public addresses), `job`, `conform`, `fidelity` (ACP against the agent's own CLI), `prove` (a benchmark's controls and proofs), `shard` and `merge` (one run over many machines), `recipe` (the builder's standard recipe), `howto` (the How to add steps) and the `cli`; its
`payload` snapshots the running backend and effective catalogue for a remote host;
conformance kits under `standard/kits/`, each a benchmark: a `driver` and, for the
tools kit, an MCP server `mcp_kit`) - the definitions it reads are in the catalogue:
Expand Down Expand Up @@ -293,6 +293,18 @@ one assistant or one benchmark.
Use `nullius.thinkflowlab@proton.me` for Nullius-owned contact email addresses,
including security reports.

### Documentation

`docs/` is terse: fragments and tables, the fewest words that change what a
reader chooses, does or concludes.

- Keep guarantees, requirements, refusals and their fix. Drop the mechanism
(flags, files, protocols, how a key travels), interface narration, history
and asides.
- A table cell is a line, not a paragraph. One fact lives on one page; others link.
- A prerequisite Review already checks is not repeated.
- `docs/standard.md` headings are never renamed: the import skills cite them.

### Attribution

**No machine attribution in commit messages or pull request descriptions** — no
Expand Down
9 changes: 7 additions & 2 deletions app/nullius/execution.py
Original file line number Diff line number Diff line change
Expand Up @@ -336,15 +336,16 @@ def withheld() -> List[str]:
names other tools give them or this application gave them before
(``settings.INSTEAD``, those that stand for a secret), the launcher's own
names for the run's key and endpoint, the agent keys (a Harbor run is
sent them by its launcher, as the key), the names a standard benchmark's
sent them by its launcher, as the key), the Boat key (a standard run on
Boat is sent it the same way), the names a standard benchmark's
own licences go by - kept in its own store (``nullius.secrets``), although
imported saved settings may still hold a value until it is moved in
Custom keys - and GitHub tokens from saved settings or the
shell Nullius was started in. Every named endpoint's settings too, so a
run gets its own endpoint's key alone. A data provider's token name is not one."""
secret = {v.name for v in settings.VARIABLES if v.secret or v.kind in ("endpoint", "provider")} | set(settings.endpoint_names())
return sorted(secret | {old for old, new in settings.INSTEAD.items() if new in secret}
| set(KEY_NAMES + URL_NAMES) | {"GITHUB_TOKEN", "GH_TOKEN"} | set(settings.agent_key_names())
| set(KEY_NAMES + URL_NAMES) | {"GITHUB_TOKEN", "GH_TOKEN", "BOAT_API_KEY"} | set(settings.agent_key_names())
| set(_licence_names()))


Expand Down Expand Up @@ -398,6 +399,10 @@ def __init__(self, plan: Plan, root: Optional[str] = None, run: Optional[Callabl
self._key, self._url = "", ""
# the agent keys Settings holds (CURSOR_API_KEY, ...): a Harbor run's agent may need its own vendor's; the standard's never gets one
self._agent_keys = dict(agent_keys) if agent_keys is not None else (settings.agent_key_values(env) if plan.runner == "harbor" else {})
if plan.runner == "standard" and (plan.recipe or {}).get("backend") == "boat":
from nullius.standard import boat # the Boat key, to the standard run that starts VMs with it, and to no other command
if boat.key(env):
self._agent_keys[boat.NAME] = boat.key(env)
# the benchmark's licences a standard run on a host is sent: read from its store at each launch, unless given
self._licences_given = licences
self._redact: List[str] = [] # the licences' values sent, as text: never in a log line
Expand Down
2 changes: 1 addition & 1 deletion app/nullius/prerequisites.py
Original file line number Diff line number Diff line change
Expand Up @@ -98,7 +98,7 @@ def check_host(host: hosts.Host, needs: List[str], setups: Optional[Dict[str, di
"""The rows for a host, from one ssh probe. Unreachable: one row, and nothing else is known."""
test = test or hosts.test
programs = [p for s in (setups or {}).values() for p in s.get("check") or []]
programs += [n for n in needs if n in ("python3", "git")]
programs += [n for n in needs if n in ("python3", "git", "ssh", "ssh-keygen", "rsync", "tar")]
got = test(host, programs=programs)
if not got["ok"]:
return [row("connect", "missing", got.get("error") or "no answer", label="Connection")]
Expand Down
Loading
Loading