Fix undeclared concept removal and prune unreachable specs

[lchoquel]

Summary

• Remove the flawed fixed_pipe_parallel_concepts optimization in _fix_undeclared_concept_references that prematurely removed concepts from the undeclared set even when referenced elsewhere, wasting a fix-loop iteration
• Add _prune_unreachable_specs method that walks the call graph from main_pipe to remove unreachable pipes and transitively unused concepts after all fixes are applied
• New method handles all controller pipe types: PipeSequence, PipeParallel, PipeBatch, and PipeCondition (filtering SpecialOutcome values)

Test plan

• make agent-check — all linting, formatting, pyright, and mypy pass
• make agent-test — full test suite passes

🤖 Generated with Claude Code

---

Note

Medium Risk
Prunes pipes/concepts in-place based on reachability and reference scanning, which could remove specs that were previously retained if graph traversal or domain detection misses an edge case.

Overview
Improves the builder fix loop by recomputing the undeclared concept set after mutating PipeParallel specs, avoiding prematurely dropping concepts that may still be referenced elsewhere.

Adds a post-fix pruning pass (_prune_unreachable_specs) that walks from main_pipe to delete unreachable pipes and then removes unused concepts (including transitive refines/structure references), with domain-aware handling so external concept refs don’t keep local concepts alive; includes new unit tests covering pruning, domain filtering, and missing-pipe warnings.

^{Written by Cursor Bugbot for commit 596c761. This will update automatically on new commits. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d28946ce9d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[lchoquel]

Code review

Found 1 issue:

1. _prune_unreachable_specs does not filter concept references by domain, unlike _collect_local_bare_concept_codes in the same PR. In Steps C and D, bare concept codes are extracted unconditionally using concept.split(".")[-1], while _collect_local_bare_concept_codes correctly checks if "." not in concept_ref_or_code or concept_ref_or_code.split(".", maxsplit=1)[0] == domain before adding. If a pipe references an external concept like external_lib.Document, the bare code Document is added to referenced_concepts, which could prevent a local unused concept with the same name from being pruned.

Step C (no domain check):

pipelex/pipelex/builder/builder_loop.py

Lines 333 to 352 in daf1843

	# Output
	output_parse = parse_concept_with_multiplicity(pipe_spec.output)
	output_concept = output_parse.concept_ref_or_code
	bare_output = output_concept.split(".")[-1] if "." in output_concept else output_concept
	referenced_concepts.add(bare_output)
	# Inputs
	if pipe_spec.inputs:
	for input_concept_str in pipe_spec.inputs.values():
	input_parse = parse_concept_with_multiplicity(input_concept_str)
	input_concept = input_parse.concept_ref_or_code
	bare_input = input_concept.split(".")[-1] if "." in input_concept else input_concept
	referenced_concepts.add(bare_input)
	# PipeParallel combined_output
	if isinstance(pipe_spec, PipeParallelSpec) and pipe_spec.combined_output:
	combined_parse = parse_concept_with_multiplicity(pipe_spec.combined_output)
	combined_concept = combined_parse.concept_ref_or_code
	bare_combined = combined_concept.split(".")[-1] if "." in combined_concept else combined_concept
	referenced_concepts.add(bare_combined)

_collect_local_bare_concept_codes (correct domain check via _add_if_local):

pipelex/pipelex/builder/builder_loop.py

Lines 414 to 427 in daf1843

	def _add_if_local(concept_ref_or_code: str) -> None:
	if "." not in concept_ref_or_code or concept_ref_or_code.split(".", maxsplit=1)[0] == domain:
	bare_code = concept_ref_or_code.rsplit(".", maxsplit=1)[-1] if "." in concept_ref_or_code else concept_ref_or_code
	referenced.add(bare_code)
	# Scan pipe specs
	if pipelex_bundle_spec.pipe:
	for pipe_spec in pipelex_bundle_spec.pipe.values():
	_add_if_local(parse_concept_with_multiplicity(pipe_spec.output).concept_ref_or_code)
	if pipe_spec.inputs:
	for input_concept_str in pipe_spec.inputs.values():
	_add_if_local(parse_concept_with_multiplicity(input_concept_str).concept_ref_or_code)
	if isinstance(pipe_spec, PipeParallelSpec) and pipe_spec.combined_output:
	_add_if_local(parse_concept_with_multiplicity(pipe_spec.combined_output).concept_ref_or_code)

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Duplicated concept-reference scanning across new methods

Low Severity

The concept-reference scanning logic (pipe outputs, inputs, PipeParallelSpec.combined_output, concept refines, structure concept_ref, and item_concept_ref) is duplicated nearly verbatim between _prune_unreachable_specs Steps C+D and _collect_local_bare_concept_codes. Both iterate the same fields and call _extract_local_bare_code identically. If a new concept-bearing field is added to any spec type, both locations need to be updated in lockstep, risking silent divergence.

Additional Locations (1)

• pipelex/builder/builder_loop.py#L438-L461