You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
|[`nemo-oo-agentdoc`](nemo-oo-agentdoc/SKILL.md)| Making types render beautiful docs for the LLM: `doc()`, `spec()`, `hidden`, `Annotated` descriptions, `pformat`/`pprint` tuning |
14
-
|[`nemo-oo-context-and-state`](nemo-oo-context-and-state/SKILL.md)| Context blocks, event history and `EventQuery`, history summarization, persistence and memory |
15
-
|[`nemo-oo-tools-and-skills`](nemo-oo-tools-and-skills/SKILL.md)| Methods as tools, built-in tools (Bash/File/Todo), MCP integration, agent skills (`Skill`/`TextSkill`), multimodal media |
|[`nooa-agentdoc`](nooa-agentdoc/SKILL.md)| Making types render beautiful docs for the LLM: `doc()`, `spec()`, `hidden`, `Annotated` descriptions, `pformat`/`pprint` tuning |
14
+
|[`nooa-context-and-state`](nooa-context-and-state/SKILL.md)| Context blocks, event history and `EventQuery`, history summarization, persistence and memory |
15
+
|[`nooa-tools-and-skills`](nooa-tools-and-skills/SKILL.md)| Methods as tools, built-in tools (Bash/File/Todo), MCP integration, agent skills (`Skill`/`TextSkill`), multimodal media |
All content was verified against `src/nooa` at the time of writing; where the skills contradict older docs (e.g. `SkillManager`, `from agentdoc import ...`, `enable_tracing(trace_dir=...)`, "private methods aren't traced"), the skills reflect the code.
24
24
@@ -29,11 +29,11 @@ Copy or symlink skill directories into the location your coding agent reads:
There are cases where `{param}` is appropriate — e.g. when prefill is disabled and you need custom rendering, or when you want to embed a short value directly in the instruction. For advanced prefill and truncation control, see `nemo-oo-codeact-advanced`.
107
+
There are cases where `{param}` is appropriate — e.g. when prefill is disabled and you need custom rendering, or when you want to embed a short value directly in the instruction. For advanced prefill and truncation control, see `nooa-codeact-advanced`.
108
108
109
109
Template expansion is primarily for values the signature *can't* show: `{self.attr}` instance state and computed expressions like `{len(items)}`. Escape literal braces as `{{ }}`.
**Constructors take `config=` only.**`PredictStrategy(max_retries=3)` and `CodeActStrategy(max_iterations=10)` are errors — wrap options in `PredictConfig(...)`/`CodeActConfig(...)`. Useful `CodeActConfig` fields: `max_iterations`, `max_retries`, `cell_timeout`, `max_tokens`, `temperature`, `max_consecutive_text_only`, `restrictions`.
160
160
161
-
`max_iterations` is a safety net, not the main tuning dial — decompose the task instead of raising the cap. For prefill control, truncation tuning, code restrictions, and the full config surface, see `nemo-oo-codeact-advanced`.
161
+
`max_iterations` is a safety net, not the main tuning dial — decompose the task instead of raising the cap. For prefill control, truncation tuning, code restrictions, and the full config surface, see `nooa-codeact-advanced`.
162
162
163
163
**Reserved parameter:** naming an agentic method parameter `reasoning` raises `ValueError` at class creation (chain-of-thought is provided via the injected `reasoning()` builtin instead).
164
164
@@ -196,8 +196,8 @@ class SearchAgent(Agent, llm=llm):
196
196
-**Hide the entry-point agentic method** (`@hidden` on `run`/`respond`) when the LLM might recursively call itself through `doc(self)`.
197
197
- Never hide a method the LLM must call as a tool.
198
198
- Module-level imports are the LLM's execution namespace: if generated code needs `json.loads`, `import json # noqa: F401` at the top of the agent file.
199
-
-`self.context` / `self.events` are always present but hidden; expose with `spec(self, "context", hidden=False)` in `__init__` (see `nemo-oo-context-and-state`).
200
-
- Making the visible surface render *well* — field descriptions, referenced-type expansion, `pformat` caps — is its own craft: see `nemo-oo-agentdoc`.
199
+
-`self.context` / `self.events` are always present but hidden; expose with `spec(self, "context", hidden=False)` in `__init__` (see `nooa-context-and-state`).
200
+
- Making the visible surface render *well* — field descriptions, referenced-type expansion, `pformat` caps — is its own craft: see `nooa-agentdoc`.
# kill -USR2 <pid> # dump traceback + all registered cells to debug_dump_<pid>.txt
224
224
```
225
225
226
-
Most bugs are visible in the rendered prompt. For runtime behavior, capture traces and inspect them — see `nemo-oo-capturing-traces` and `nemo-oo-trace-explorer`.
226
+
Most bugs are visible in the rendered prompt. For runtime behavior, capture traces and inspect them — see `nooa-capturing-traces` and `nooa-trace-explorer`.
Copy file name to clipboardExpand all lines: skills/nooa-agentdoc/SKILL.md
+6-6Lines changed: 6 additions & 6 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,7 +1,7 @@
1
1
---
2
-
name: nemo-oo-agentdoc
2
+
name: nooa-agentdoc
3
3
description: Make NVIDIA OO agent types render beautiful documentation for the LLM — doc(), spec(), hidden, Annotated field descriptions, and pformat/pprint tuning. Use when designing Pydantic models/dataclasses the LLM will see, controlling what appears in doc(self), hiding internals, adding field descriptions, fixing noisy or missing type docs, or tuning value truncation.
4
-
compatibility: nemo-oo-agents core package (agentdoc ships inside it — import from nooa.agentdoc)
4
+
compatibility: nooa package (agentdoc ships inside it — import from nooa.agentdoc)
- Classes with their own `__repr__` (pandas, numpy) are trusted and rendered via their repr (truncated head/tail if huge).
141
141
- Implement `__instance_values__(self) -> dict[str, Any]` (the `SupportsInstanceValues` protocol) to control exactly which fields an instance renders — omitted keys are hidden.
142
-
- These same knobs are what `TruncationConfig`'s `event_format`/`prefill_format` splat into `pformat` framework-wide — see `nemo-oo-codeact-advanced`.
142
+
- These same knobs are what `TruncationConfig`'s `event_format`/`prefill_format` splat into `pformat` framework-wide — see `nooa-codeact-advanced`.
Copy file name to clipboardExpand all lines: skills/nooa-capturing-traces/SKILL.md
+5-5Lines changed: 5 additions & 5 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,7 +1,7 @@
1
1
---
2
-
name: nemo-oo-capturing-traces
2
+
name: nooa-capturing-traces
3
3
description: Capture execution traces from NVIDIA OO Agents. Use when instrumenting an agent run, writing traces to JSONL files, sending traces to the viewer or an OTLP/Langfuse/Phoenix backend, controlling which methods are traced, or when traces are mysteriously missing.
4
-
compatibility: nemo-oo-agents core package; the [tracing] extra (opentelemetry + openinference) for exporters
4
+
compatibility: nooa package; the [tracing] extra (opentelemetry + openinference) for exporters
5
5
---
6
6
7
7
# Capturing Traces
@@ -105,7 +105,7 @@ Files are OTLP JSON Lines: each line is one `{"resourceSpans": [...]}` object. T
105
105
106
106
## Pitfalls
107
107
108
-
-**Do NOT use**`from openinference_instrumentation_nemo_oo_agents import enable_tracing` or `enable_tracing(trace_dir=...)` — both appear in older docs/comments but do not exist. Tracing lives in `nooa.tracing`; `trace_dir` is an argument of `exporters.jsonl()`, and `enable_tracing()` returns `None`.
108
+
-**Do NOT use**legacy OpenInference instrumentation imports or `enable_tracing(trace_dir=...)` — both appear in older docs/comments but do not exist. Tracing lives in `nooa.tracing`; `trace_dir` is an argument of `exporters.jsonl()`, and `enable_tracing()` returns `None`.
109
109
- Auto-tracing is attempted once per process. If the viewer wasn't running when the first `Agent` was constructed, later agents won't retry — call `enable_tracing(...)` explicitly or restart with the viewer up.
110
110
- For short scripts, call `flush_traces()` before exit; batch exporters flush on a ~1s schedule and a fast exit can drop the tail of a trace.
111
111
- Logic that runs *outside* agent methods (module-level preprocessing, `main()` helpers) is invisible in traces. Keep interesting logic inside agent methods so failures leave trace evidence.
@@ -123,5 +123,5 @@ flush_traces()
123
123
124
124
## Related skills
125
125
126
-
-`nemo-oo-trace-viewer` — run the web viewer and browse captured traces.
127
-
-`nemo-oo-trace-explorer` — programmatic/CLI trace analysis and root-cause debugging.
126
+
-`nooa-trace-viewer` — run the web viewer and browse captured traces.
127
+
-`nooa-trace-explorer` — programmatic/CLI trace analysis and root-cause debugging.
Copy file name to clipboardExpand all lines: skills/nooa-channels/SKILL.md
+6-6Lines changed: 6 additions & 6 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,7 +1,7 @@
1
1
---
2
-
name: nemo-oo-channels
2
+
name: nooa-channels
3
3
description: Reactive input for NVIDIA OO agents — Channel/QueueManager for queued and event-mode input, race() turn dispatch, spawn() background jobs with JobHandle, and the bundled producers (monitor a shell command, cron ticks, one-shot timers, file tails). Use when an agent must react to external input mid-run — user messages, CI output, timers, job completions — or when building an interactive/long-running agent loop.
-`QueueManager` is hidden from the LLM by default; expose deliberately (`spec(self, "queue_manager", hidden=False)`) or, better, expose only the `reader` attributes and keep `put`/`spawn` on the Python side.
77
77
- Registration order is priority order for `race()` — register the highest-priority channel (usually user messages) first.
78
78
- Don't `await channel.get()` (producer object) from LLM code — give the LLM the `.reader`, whose `get(timeout=...)` can't deadlock a cell.
79
-
- The dispatcher belongs in a pure-Python orchestrator method (`run()` above) — see "Orchestrators are pure Python" in `nemo-oo-agent-authoring`.
79
+
- The dispatcher belongs in a pure-Python orchestrator method (`run()` above) — see "Orchestrators are pure Python" in `nooa-agent-authoring`.
80
80
-`spawn()` requires the channel to already exist (`ValueError` otherwise).
81
81
82
82
## Related skills
83
83
84
-
-`nemo-oo-agent-authoring` — the orchestrator pattern the dispatch loop lives in.
85
-
-`nemo-oo-context-and-state` — `QueueOutput`/`StreamEnd`/`JobError` are events; query them like any others.
86
-
-`nemo-oo-middleware-hooks` — `on()` observers if you only need to react to recorded events, not consume input.
84
+
-`nooa-agent-authoring` — the orchestrator pattern the dispatch loop lives in.
85
+
-`nooa-context-and-state` — `QueueOutput`/`StreamEnd`/`JobError` are events; query them like any others.
86
+
-`nooa-middleware-hooks` — `on()` observers if you only need to react to recorded events, not consume input.
Copy file name to clipboardExpand all lines: skills/nooa-codeact-advanced/SKILL.md
+6-6Lines changed: 6 additions & 6 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,12 +1,12 @@
1
1
---
2
-
name: nemo-oo-codeact-advanced
2
+
name: nooa-codeact-advanced
3
3
description: Advanced tuning of NVIDIA OO Agents strategies — CodeAct prefill (understanding, disabling, custom, pre-ellipsis code), loop guards (max_iterations, retries, text-only stop), truncation tuning (TruncationConfig/CaptureConfig/FormatConfig), code restrictions (RestrictionsConfig), execution environment internals, and PredictStrategy tuning (retries, param guards, output_serialization). Use when configuring CodeActConfig or PredictConfig beyond defaults, writing a custom prefill, restricting generated code, or debugging truncation/eviction behavior.
4
-
compatibility: nemo-oo-agents core package
4
+
compatibility: nooa package
5
5
---
6
6
7
7
# Advanced CodeAct (and Predict) Tuning
8
8
9
-
The authoring basics are in `nemo-oo-agent-authoring`. This skill covers the deep configuration surface, verified against `strategies/codeact.py`, `strategies/prefill.py`, `strategies/predict.py`, and `config/`.
9
+
The authoring basics are in `nooa-agent-authoring`. This skill covers the deep configuration surface, verified against `strategies/codeact.py`, `strategies/prefill.py`, `strategies/predict.py`, and `config/`.
10
10
11
11
## Config plumbing rules (read first)
12
12
@@ -124,6 +124,6 @@ Single LLM turn, no tools, no code. The prompt is the docstring plus each parame
124
124
125
125
## Related skills
126
126
127
-
-`nemo-oo-agent-authoring` — the basics this builds on (strategy selection, contracts, visibility).
128
-
-`nemo-oo-context-and-state` — the context blocks that truncation/eviction act on.
129
-
-`nemo-oo-capturing-traces` / `nemo-oo-trace-explorer` — see every prefill cell, tool call, and validation retry in the trace.
127
+
-`nooa-agent-authoring` — the basics this builds on (strategy selection, contracts, visibility).
128
+
-`nooa-context-and-state` — the context blocks that truncation/eviction act on.
129
+
-`nooa-capturing-traces` / `nooa-trace-explorer` — see every prefill cell, tool call, and validation retry in the trace.
0 commit comments