swarm-notes 0.3.3__tar.gz → 0.3.4__tar.gz

{swarm_notes-0.3.3 → swarm_notes-0.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: swarm-notes
-Version: 0.3.3
+Version: 0.3.4
 Summary: Automated research paper tracking and knowledge synthesis
 Author: LM
 Author-email: LM <hi@leima.is>
@@ -237,4 +237,12 @@ swarm-notes run \
 
 ## License
 
-MIT — see [LICENSE](LICENSE).
+MIT — see [LICENSE](LICENSE).
+## Releasing
+
+Releases are published to PyPI by `.github/workflows/release.yml` when a `v*` tag is pushed:
+
+1. Bump `version` in `pyproject.toml` and merge to `main`.
+2. Tag the merge commit with the same version, e.g. `git tag v0.3.4 && git push origin v0.3.4`.
+
+The workflow checks that the tag matches `pyproject.toml`, runs the tests, builds, and publishes with PyPI trusted publishing. One-time setup: add this repository as a trusted publisher for `swarm-notes` on PyPI (workflow `release.yml`, environment `pypi`).

{swarm_notes-0.3.3 → swarm_notes-0.3.4}/README.md

@@ -218,4 +218,12 @@ swarm-notes run \
 
 ## License
 
-MIT — see [LICENSE](LICENSE).
+MIT — see [LICENSE](LICENSE).
+## Releasing
+
+Releases are published to PyPI by `.github/workflows/release.yml` when a `v*` tag is pushed:
+
+1. Bump `version` in `pyproject.toml` and merge to `main`.
+2. Tag the merge commit with the same version, e.g. `git tag v0.3.4 && git push origin v0.3.4`.
+
+The workflow checks that the tag matches `pyproject.toml`, runs the tests, builds, and publishes with PyPI trusted publishing. One-time setup: add this repository as a trusted publisher for `swarm-notes` on PyPI (workflow `release.yml`, environment `pypi`).

{swarm_notes-0.3.3 → swarm_notes-0.3.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "swarm-notes"
-version = "0.3.3"
+version = "0.3.4"
 description = "Automated research paper tracking and knowledge synthesis"
 readme = "README.md"
 authors = [

{swarm_notes-0.3.3 → swarm_notes-0.3.4}/src/swarm_notes/analyst.py

@@ -41,21 +41,21 @@ class ConceptLink(BaseModel):
         description="One sentence that defines this concept in the context of the paper."
     )
     importance_reason: str = Field(
-        # default="",
+        default="",
         description=(
             "Why this concept is central to the paper's contribution or novelty. "
             "Leave empty if not applicable."
         ),
     )
     reusability_reason: str = Field(
-        # default="",
+        default="",
         description=(
             "Why this concept is likely to recur across future papers and deserves "
             "a permanent vault note. Leave empty if not applicable."
         ),
     )
     evidence_excerpt: str = Field(
-        # default="",
+        default="",
         description=(
             "Short phrase or sentence from the title or abstract supporting this concept. "
             "Leave empty if not applicable."
@@ -79,14 +79,14 @@ class OpenQuestion(BaseModel):
         description="Detailed explanation of the open question or future work."
     )
     importance_reason: str = Field(
-        # default="",
+        default="",
         description=(
             "Why this open question is technically important and worth tracking across papers. "
             "Leave empty if not applicable."
         ),
     )
     evidence_excerpt: str = Field(
-        # default="",
+        default="",
         description=(
             "Short phrase or sentence from the paper text supporting that this question is explicit and unresolved. "
             "Leave empty if not applicable."
@@ -155,7 +155,7 @@ class PaperAnalysisBase(BaseModel):
     )
 
     architectures: list[str] = Field(
-        # default_factory=list,
+        default_factory=list,
         description=(
             "Architecture names used or proposed (from the taxonomy architectures list). "
             "Leave empty if not applicable."
@@ -163,7 +163,7 @@ class PaperAnalysisBase(BaseModel):
     )
 
     datasets: list[DatasetLink] = Field(
-        # default_factory=list,
+        default_factory=list,
         description=(
             "Names of CRITICAL datasets used for evaluation or training "
             "(e.g. 'ImageNet', 'ETTh1', 'LAION-5B'). "
@@ -172,7 +172,7 @@ class PaperAnalysisBase(BaseModel):
     )
 
     concepts: list[ConceptLink] = Field(
-        # default_factory=list,
+        default_factory=list,
         description=(
             "Key concepts introduced or heavily used by the paper that "
             "should have their own vault entry (e.g. a new architecture, "
@@ -181,7 +181,7 @@ class PaperAnalysisBase(BaseModel):
     )
 
     limitations: str = Field(
-        # default="",
+        default="",
         description="Brief statement of known limitations or future work.",
     )
 
@@ -193,34 +193,54 @@ class PaperAnalysisBase(BaseModel):
     )
 
     open_questions: list[OpenQuestion] = Field(
-        # default_factory=list,
+        default_factory=list,
         description="List of open questions or future research identified by the domain expert.",
     )
 
     critic_review_summary: str = Field(
-        # default="",
+        default="",
         description="Short archivist review summary explaining why items were accepted or rejected.",
     )
 
     critic_rejected_candidates: list[RejectedCandidate] = Field(
-        # default_factory=list,
+        default_factory=list,
         description="Structured rejection records for candidate concepts or open questions rejected by the archivist review.",
     )
 
 
 class PaperAnalysis(PaperAnalysisBase):
     """Structured extraction output for a single academic paper."""
-    tags: list[str] = Field(default_factory=list, description="All assigned tags.")
+    tags: list[str] = Field(
+        default_factory=list,
+        description=(
+            "Tags for this paper. Reuse tags from the EXISTING TAGS list whenever one applies; "
+            "only add a new tag if no existing tag describes the idea. Maximum 10 tags."
+        ),
+    )
+
+
+# Upper bound on how many existing tags are listed in the prompt. Tags are
+# passed as prompt text rather than as an enum in the output schema: a schema
+# enum with hundreds of values makes Gemini reject the request with
+# 400 INVALID_ARGUMENT.
+MAX_PROMPT_TAGS = 1000
 
 
 # ---------------------------------------------------------------------------
 # Agent factory
 # ---------------------------------------------------------------------------
 
-def _build_system_prompt(skill: SkillSpec, taxonomy: dict) -> str:
+def _build_system_prompt(
+    skill: SkillSpec, taxonomy: dict, existing_tags: list[str] | None = None
+) -> str:
     architectures_list = ", ".join(taxonomy.get("architectures", []))
     domains_list = ", ".join(taxonomy.get("domains", []))
-    
+    existing_tags_str = (
+        ", ".join(existing_tags[:MAX_PROMPT_TAGS])
+        if existing_tags
+        else "(No tags currently exist in the vault)"
+    )
+
     base_prompt = f"""You are an expert academic paper analyst specialising in machine learning research.
 
 Your task is to analyse the provided paper abstract and extract structured information.
@@ -229,11 +249,14 @@ TAXONOMY (use ONLY these values for architectures/domains):
 - Available architectures: {architectures_list}
 - Available domains: {domains_list}
 
+EXISTING TAGS IN VAULT:
+{existing_tags_str}
+
 SKILL CONTEXT ({skill.name}):
 {skill.extra_system_prompt}
 
 IMPORTANT RULES:
-1. Select tags according to the dynamic schema.
+1. Prefer tags from the EXISTING TAGS list, spelled exactly as listed. Add a new tag only when none applies, and keep new tags to an absolute minimum.
 2. Select domain ONLY from the taxonomy domains list above.
 3. Select architectures ONLY from the taxonomy architectures list above (or leave empty).
 4. The summary must be 3-5 sentences, technically precise but accessible.
@@ -268,46 +291,12 @@ def analyse(paper: RawPaper, skill: SkillSpec) -> PaperAnalysis:
         Structured extraction result.
     """
     taxonomy = _load_taxonomy()
-    system_prompt = _build_system_prompt(skill, taxonomy)
     existing_tags = get_existing_tags()
-    
-    from pydantic import create_model
-
-    fields = {}
-    if existing_tags:
-        ExistingTags = Literal[tuple(existing_tags)]  # type: ignore
-        fields["existing_tags"] = (
-            list[ExistingTags],
-            Field(
-                default_factory=list,
-                description="Tags chosen from the existing knowledge base. Pick only tags that genuinely apply."
-            )
-        )
-        fields["new_tags"] = (
-            list[str],
-            Field(
-                default_factory=list,
-                description="Use this ONLY if the concept cannot be accurately described by any of the existing_tags. Keep new tags to an absolute minimum."
-            )
-        )
-    else:
-        fields["new_tags"] = (
-            list[str],
-            Field(
-                default_factory=list,
-                description="List of new tags for this paper."
-            )
-        )
-        
-    DynamicPaperAnalysis = create_model(
-        "DynamicPaperAnalysis",
-        __base__=PaperAnalysisBase,
-        **fields
-    )
+    system_prompt = _build_system_prompt(skill, taxonomy, existing_tags)
 
-    agent: Agent[None, DynamicPaperAnalysis] = Agent(
+    agent: Agent[None, PaperAnalysis] = Agent(
         model=settings.llm_model,
-        output_type=DynamicPaperAnalysis,
+        output_type=PaperAnalysis,
         system_prompt=system_prompt,
     )
 
@@ -323,17 +312,8 @@ def analyse(paper: RawPaper, skill: SkillSpec) -> PaperAnalysis:
     logger.info("Analyst: analysing paper %s with skill %s", paper.arxiv_id, skill.name)
 
     result = agent.run_sync(user_message)
-    dynamic_analysis = result.output
-    
-    # Merge existing and new tags
-    combined_tags = []
-    if hasattr(dynamic_analysis, "existing_tags"):
-        combined_tags.extend(getattr(dynamic_analysis, "existing_tags", []))
-    if hasattr(dynamic_analysis, "new_tags"):
-        combined_tags.extend(getattr(dynamic_analysis, "new_tags", []))
-        
-    analysis_dict = dynamic_analysis.model_dump(exclude={"existing_tags", "new_tags"})
-    analysis = PaperAnalysis(**analysis_dict, tags=combined_tags)
+    analysis: PaperAnalysis = result.output
+    analysis.tags = _normalise_tags(analysis.tags, existing_tags)
 
     # Ensure metadata from the raw paper is preserved accurately
     analysis.arxiv_id = paper.arxiv_id
@@ -352,6 +332,22 @@ def analyse(paper: RawPaper, skill: SkillSpec) -> PaperAnalysis:
     return analysis
 
 
+def _normalise_tags(tags: list[str], existing_tags: list[str]) -> list[str]:
+    """Map tags onto the vault's existing spelling (case-insensitive) and drop duplicates."""
+    canonical = {t.lower(): t for t in existing_tags}
+    result: list[str] = []
+    seen: set[str] = set()
+    for tag in tags:
+        tag = tag.strip()
+        if not tag:
+            continue
+        tag = canonical.get(tag.lower(), tag)
+        if tag.lower() not in seen:
+            seen.add(tag.lower())
+            result.append(tag)
+    return result
+
+
 def _load_taxonomy() -> dict:
     """Load taxonomy.json, returning an empty dict on failure."""
     if not settings.taxonomy_file.exists():

{swarm_notes-0.3.3 → swarm_notes-0.3.4}/src/swarm_notes/main.py

@@ -328,6 +328,13 @@ def run(
             len(papers),
         )
 
+        if not analyses:
+            # Every paper failed (e.g. the LLM rejected every request). Exit
+            # non-zero so the scheduled workflow shows up as failed instead of
+            # silently committing nothing.
+            logger.error("No papers were processed successfully – failing the run")
+            raise typer.Exit(1)
+
     except typer.Exit:
         # Re-raise Typer exit as-is
         raise

{swarm_notes-0.3.3 → swarm_notes-0.3.4}/src/swarm_notes/vault_writer.py

@@ -616,5 +616,14 @@ def _slugify(text: str) -> str:
 
 
 def _yaml_escape(text: str) -> str:
-    """Escape double quotes in YAML string values."""
-    return text.replace('"', '\\"')
+    """Escape text for use inside a double-quoted YAML scalar.
+
+    Backslashes must be escaped first: titles with LaTeX such as
+    ``$\\mathrm{SO}(2)$`` otherwise produce invalid escape sequences and
+    break every YAML parser that reads the note (including the site build).
+    """
+    return (
+        text.replace("\\", "\\\\")
+        .replace('"', '\\"')
+        .replace("\n", " ")
+    )