exo/lm-lib
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

docs: Clarify execution graph legend in README with detailed symbol meanings and example trace

Browse Source
This commit is contained in:
extio-de
2025-10-01 19:59:59 +02:00
parent 89c0cc5867
commit b638d57047
1 changed files with 19 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
README.md
View File

@@ -246,21 +246,28 @@ Utility helper supplied: `BaseAgent.mergeContexts(keys, contexts)` merges list-v
## Execution Graph Legend ## Execution Graph Legend
The `graph` in `AgentContext` accumulates symbols documenting flow: The `graph` in `AgentContext` accumulates symbols documenting the execution flow. These symbols provide a visual trace of the agent pipeline, showing model usage, branching/merging, and any issues encountered:
| Symbol | Meaning | | Symbol | Meaning |
|--------|---------| |--------|---------|
| `M` / `S` / `L` / ... | Model category short name | | `M` / `S` / `L` / `HOT` / `COLD` | Model category short name (Medium, Small, Large, Hot, Cold) |
| Agent name | Executed agent identifier | | Agent name | Name of the executed agent (e.g., `CodeSummarizer`, `BlogPostAgent`) |
| `○` | Completion skipped (`skipNextCompletion` or processing-only) | | `○` | Completion skipped (either `skipNextCompletion` flag or `PROCESSING_ONLY` agent type) |
| `⇉ x/n` | Branch split index x of n | | `⇉ x/n` | Branch split index x of n (parallel execution from template placeholder branching) |
| `size↣merged` | Result of merge phase (e.g., `5↣1`) | | `size↣merged` | Result of merge phase showing input→output count (e.g., `5↣1` means 5 contexts merged into 1) |
| `→` | Next agent of type COMPLETION | | `→` | Next agent is of type `COMPLETION` (stateless single-shot, conversation cleared) |
| `🗨` | Next agent starts new conversation | | `🗨` | Next agent starts a new conversation (conversation history cleared) |
| `↴` | Continue conversation (assistant turn appended if missing) | | `↴` | Continue existing conversation (assistant turn appended if missing, maintains context) |
| `✓` / `✗` | Grading pass/fail (if `AgentNext.gradingPassed()` present) | | `✓` / `✗` | Grading result (pass/fail) when `AgentNext.gradingPassed()` is present |
| `⚠` | Parse attempt failed (retry followed) | | `⚠` | Parse attempt failed (retry will follow, max 2 attempts total) |
| `☢` | Fatal parse failure / unknown agent / not found | | `☢` | Fatal error: parse failure after all retries, unknown agent, or agent not found |
| `‖` | Agent execution continuation (appears when calling `walk()` on a non-empty graph) |
**Example graph trace:**
```
M, CodeSummarizer, ↴, M, FeatureIdentifier, →, ⇉ 1/5, M, FeatureAnalyzer, 5↣1, →, M, AnswerGenerator
```
This shows: Medium model used for CodeSummarizer → conversation continued to FeatureIdentifier → branches into 5 parallel splits → each split analyzed with FeatureAnalyzer → merged back to 1 context → final answer generated.
## Retry & Parse Handling ## Retry & Parse Handling