Module: vvaharness/pipeline

docs / modules / pipeline

← all docs

Module: vvaharness/pipeline

Role. The assembly line. Each stage transforms the previous stage's artifact, from static seed to SARIF report, plus the tree-sitter call-graph engine that feeds them.

Size. 18 files · ~14,185 LOC.


flowchart LR IN["Input\nTarget repo (+ optional static seed)"] --> ACT["Action\nRun the S0–S9 detection stages against the target"] ACT --> VAL["Validation\nS5 deterministic confidence/evidence gates · S6 adversarial verification"] VAL --> OUT["Output\nFinalReport → report.md + report.sarif"]

This is the heart of VVAH — the nine detection stages plus the static call-graph engine. Code goes in; a ranked, evidence-backed report comes out. The stages are ordered and each has a single job: map, threat-model, plan, deep-dive, filter, verify, dedup, chain, and emit. The callgraph_engine is a language-aware static analyser (tree-sitter) that traces how untrusted data could reach dangerous operations, focusing the AI where it matters.

Explain-it-simply. The conveyor belt of specialists. Each station does one inspection and passes the part along.

Data-flow (this module)

flowchart LR R[(repo)] --> S0[s0_seed] S0 --> S1[s1_preprocess] --> S2[s2_threatmodel] --> S3[s3_decompose] S3 --> S4[s4_deepdive] --> S5[s5_prefilter] --> S6[s6_verify] S6 --> S7[s7_dedup] --> S8[s8_chain] --> S9[s9 SARIF] CE[callgraph_engine] -.->|seed + taint facts| S0 CC[callgraph_consumer] -.->|graph views| S3 & S5 & S6 & S7 & S8

Technical details

flowchart LR IN["Input\nTarget repo (+ optional static seed)"] --> ACT["Action\nRun the S0–S9 detection stages against the target"] ACT --> VAL["Validation\nS5 deterministic confidence/evidence gates · S6 adversarial verification"] VAL --> OUT["Output\nFinalReport → report.md + report.sarif"]

Each sN_*.py file exposes a run(...) entry consumed by orchestrator/scan.py. s1_preprocess builds the ContextPackage; s2_threatmodel the ThreatModel; s3_decompose the TaskManifest; s4_deepdive runs specialist lenses per chunk (with optional N-run voting); s5_prefilter applies deterministic gates (plus an optional semantic pre-dedup); s6_verify is adversarial TRUE/FP + CVSS; s7_dedup deterministic+semantic dedup; s8_chain builds exploit chains and re-ranks into a FinalReport; s9 is deterministic SARIF (implemented in the report layer). s1_autoexclude derives a per-target exclusion overlay; s11_validate is a thin wrapper hooking the validation package into the pipeline. The callgraph_engine/ sub-package is the static core: _scan.py (~2,900 LOC) is the per-file tree-sitter scanner emitting FileIndex (calls, taint facts, reflection, framework markers); _graph.py (~2,080 LOC) builds the interprocedural graph and source→sink reachability; _rules.py loads source/sink/sanitizer specs; _annotator.py is optional LLM spec derivation. callgraph_consumer.py provides shared read-only graph views (GraphView) to S3 and S5–S8.

File map (quick scan)

File LOC Purpose
pipeline/__init__.py 15 —
pipeline/callgraph_consumer.py 554 Shared post-S2 callgraph consumers.
pipeline/stages/callgraph_engine/__init__.py 262 Tree-sitter-native call-graph seed engine.
pipeline/stages/callgraph_engine/_annotator.py 452 LLM-backed source/sink spec derivation for callgraph mode.
pipeline/stages/callgraph_engine/_graph.py 2077 Naive call graph + source→sink reachability.
pipeline/stages/callgraph_engine/_rules.py 450 Convert sources.generated.yaml / sinks.generated.yaml (semgrep-format
pipeline/stages/callgraph_engine/_scan.py 2911 Per-file tree-sitter scanner. Emits a :class:FileIndex containing:
pipeline/stages/s0_seed.py 277 Step 0 — Static seed.
pipeline/stages/s11_validate.py 56 s11_validate — thin wrapper that hooks the validation package into the pipeline.
pipeline/stages/s1_autoexclude.py 368 Auto-derive a per-target step1 exclusion overlay.
pipeline/stages/s1_preprocess.py 1548 Step 1 — Claude CLI explores the repo agentically and emits a ContextPackage.
pipeline/stages/s2_threatmodel.py 611 Step 2 — Application threat model.
pipeline/stages/s3_decompose.py 1521 Step 3 — the strategist LLM receives the ContextPackage (no raw code) and produces a
pipeline/stages/s4_deepdive.py 1518 Step 4 — Deep-dive each chunk N times, then majority-vote.
pipeline/stages/s5_prefilter.py 249 Step 5 — Deterministic pre-filter.
pipeline/stages/s6_verify.py 426 Step 6 — Adversarial verification.
pipeline/stages/s7_dedup.py 371 Step 7 — Deduplication.
pipeline/stages/s8_chain.py 519 Step 8 — the chain-analysis LLM sees ALL findings together and:

Files & functions, explained

Every class and function below has a one-line plain-English note — readable whether you write code or not. _name marks an internal helper.

pipeline/__init__.py

15 LOC. Utility / support module.

No top-level classes or functions (package init / constants).

pipeline/callgraph_consumer.py

554 LOC. Shared post-S2 callgraph consumers.

Classes - GraphView — Represents the “GraphView” data/behaviour used in this file.

Functions - _q_file() — (internal helper) Provides the “q file” logic. - _q_name() — (internal helper) Provides the “q name” logic. - parse_hop() — Reads and cleans up the hop value. - _norm_file() — (internal helper) reads and cleans up the file value. - _file_from_ref() — (internal helper) Provides the “file from ref” logic. - _evidence_touched_files() — (internal helper) Provides the “evidence changed files” logic. - seed_reachable_files() — Provides the “seed reachable files” logic. - seed_paths_by_file() — Provides the “seed paths by file” logic. - best_source_from_seed() — Provides the “best untrusted input point from seed” logic. - seed_evidence_reachable_files() — Provides the “seed evidence reachable files” logic. - seed_evidence_by_file() — Provides the “seed evidence by file” logic. - best_source_from_evidence() — Provides the “best untrusted input point from evidence” logic. - entry_anchor_lines() — Provides the “entry anchor lines” logic. - build_span_index() — Builds / writes the span index step. - graph_qnodes_by_file() — Provides the “code relationship map nodes in the code graph by file” logic. - reverse_edges() — Provides the “reverse edges” logic. - qnodes_at() — Provides the “nodes in the code graph at” logic. - neighborhood() — Provides the “neighborhood” logic. - graph_view() — Provides the “code relationship map view” logic. - cfgs_for_function() — Provides the “cfgs for function” logic. - reflection_facts_for_function() — Provides the “dynamic code lookup facts for function” logic. - reflection_facts_by_file() — Provides the “dynamic code lookup facts by file” logic. - condition_edges_in_evidence() — Provides the “condition edges in evidence” logic. - reflection_edges_in_evidence() — Provides the “dynamic code lookup edges in evidence” logic. - framework_markers_for_function() — Provides the “web framework markers for function” logic. - route_facts_for_file() — Provides the “web URL route facts for file” logic. - response_dataflows_for_function() — Provides the “response dataflows for function” logic.

pipeline/stages/callgraph_engine/__init__.py

262 LOC. Tree-sitter-native call-graph seed engine.

Functions - run() — Runs the run value. - _lang_of() — (internal helper) Provides the “lang of” logic. - _scan_repo() — (internal helper) runs the code repository value.

pipeline/stages/callgraph_engine/_annotator.py

452 LOC. LLM-backed source/sink spec derivation for callgraph mode.

Classes - _Candidate — (internal) Represents the “Candidate” data/behaviour used in this file.

Functions - _as_float() — (internal helper) Provides the “as float” logic. - _cfg_get() — (internal helper) Provides the “code control-flow get” logic. - _norm_cwe() — (internal helper) reads and cleans up the weakness type value. - _semantic_family() — (internal helper) Provides the “semantic family” logic. - _owasp_labels() — (internal helper) Provides the “owasp labels” logic. - _collect_candidates() — (internal helper) looks up / reads the candidates value. - _build_prompt_batch() — (internal helper) builds / writes the prompt batch step. - _parse_results() — (internal helper) reads and cleans up the results value. - _tok() — (internal helper) Provides the “token/cost estimate” logic. - _heuristic_source_kind() — (internal helper) Provides the “heuristic untrusted input point kind” logic. - _heuristic_sink_kind() — (internal helper) Provides the “heuristic dangerous operation kind” logic. - _append_spec() — (internal helper) adds detail to the rule specification value. - _seed_seen_sig() — (internal helper) Provides the “seed seen sig” logic. - supplement_with_heuristics() — Provides the “supplement with heuristics” logic. - detect_specs() — Looks up / reads the specs value.

pipeline/stages/callgraph_engine/_graph.py

2077 LOC. Naive call graph + source→sink reachability.

Functions - _score_path() — (internal helper) checks / decides the path value. - _extract_sink_family() — (internal helper) reads and cleans up the dangerous operation family step. - _effective_global_budget() — (internal helper) Provides the “effective global spending limit” logic. - _filter_paths_by_budget() — (internal helper) checks / decides the paths by spending limit step. - _cwe_num() — (internal helper) Provides the “weakness type num” logic. - _is_high_risk() — (internal helper) checks / decides the high risk step. - _prefix_score() — (internal helper) Provides the “prefix score” logic. - _compatible_taint_pair() — (internal helper) Provides the “compatible untrusted-data flow pair” logic. - _resolve_called_fids() — (internal helper) checks / decides the called fids step. - _edge_confidence() — (internal helper) Provides the “edge confidence” logic. - _fn_id() — (internal helper) Provides the “fn identifier” logic. - _fid_name() — (internal helper) Provides the “fid name” logic. - _param_symbol() — (internal helper) Provides the “param symbol” logic. - _fact_fid() — (internal helper) Provides the “fact fid” logic. - _call_targets_for_fact() — (internal helper) Provides the “call targets for fact” logic. - _has_phase1_facts_for_path() — (internal helper) checks / decides the phase1 facts for path step. - _find_matching_call_facts() — (internal helper) looks up / reads the matching call facts step. - _seed_source_taint() — (internal helper) Provides the “seed untrusted input point untrusted-data flow” logic. - _apply_field_writes() — (internal helper) runs the field writes step. - _apply_field_reads() — (internal helper) runs the field reads step. - _apply_container_writes() — (internal helper) runs the container writes step. - _apply_sanitizers() — (internal helper) runs the sanitizers value. - _apply_local_aliases() — (internal helper) runs the local aliases step. - _call_tainted_arg_slots() — (internal helper) Provides the “call tainted arg slots” logic. - _callee_may_return_tainted() — (internal helper) Provides the “callee may return tainted” logic. - _apply_return_to_local() — (internal helper) runs the return to local step. - _transfer_closure() — (internal helper) Provides the “transfer closure” logic. - _build_taint_evidence_for_path() — (internal helper) builds / writes the untrusted-data flow evidence for path step. - _apply_condition_taint() — (internal helper) runs the condition untrusted-data flow step. - _condition_edge_confidence() — (internal helper) Provides the “condition edge confidence” logic. - _solve_with_cfgs() — (internal helper) Provides the “solve with cfgs” logic. - _resolve_reflection_targets() — (internal helper) checks / decides the dynamic code lookup targets step. - _emit_reflection_taint_edges() — (internal helper) builds / writes the dynamic code lookup untrusted-data flow edges step. - _apply_reflection_to_taint() — (internal helper) runs the dynamic code lookup to untrusted-data flow step. - _build_symbol_table() — (internal helper) builds / writes the symbol table step. - _build_framework_symbol_table() — (internal helper) builds / writes the web framework symbol table step. - _apply_framework_markers() — (internal helper) runs the web framework markers step. - _emit_framework_entry_points() — (internal helper) builds / writes the web framework entry points step. - _apply_response_dataflow() — (internal helper) runs the response data movement step. - build_taint_paths() — Builds / writes the untrusted-data flow paths step.

pipeline/stages/callgraph_engine/_rules.py

450 LOC. Convert sources.generated.yaml / sinks.generated.yaml (semgrep-format

Classes - MatchSpec — Represents the “MatchSpec” data/behaviour used in this file. - has_qualified() — Checks / decides the qualified value. - has_module_attr() — Checks / decides the module attr step.

Functions - _normalize_family() — (internal helper) reads and cleans up the family value. - _owasp_2025_labels() — (internal helper) Provides the “owasp 2025 labels” logic. - _norm_langs() — (internal helper) reads and cleans up the langs value. - _extract_methods_from_regex() — (internal helper) reads and cleans up the methods from regex step. - _metavar_methods() — (internal helper) Provides the “metavar methods” logic. - _top_pattern_string() — (internal helper) Provides the “highest-priority pattern string” logic. - _walk_pattern_leaves() — (internal helper) Provides the “walk pattern leaves” logic. - _walk_rule_keys() — (internal helper) Provides the “walk rule keys” logic. - _translation_flags() — (internal helper) Provides the “translation flags” logic. - _try_module_attr() — (internal helper) Provides the “try module attr” logic. - _spec_from_rule() — (internal helper) Provides the “rule specification from rule” logic. - _load() — (internal helper) looks up / reads the load value. - load_rulepacks() — Looks up / reads the rulepacks value.

pipeline/stages/callgraph_engine/_scan.py

2911 LOC. Per-file tree-sitter scanner. Emits a :class:FileIndex containing:

Classes - CallSite — Represents the “CallSite” data/behaviour used in this file. - FuncDef — Represents the “FuncDef” data/behaviour used in this file. - ObservedCall — Represents the “ObservedCall” data/behaviour used in this file. - VarAssignFact — Represents the “VarAssignFact” data/behaviour used in this file. - ReturnFact — Represents the “ReturnFact” data/behaviour used in this file. - CallArgFact — Represents the “CallArgFact” data/behaviour used in this file. - FieldWriteFact — Represents the “FieldWriteFact” data/behaviour used in this file. - FieldReadFact — Represents the “FieldReadFact” data/behaviour used in this file. - ContainerWriteFact — Represents the “ContainerWriteFact” data/behaviour used in this file. - FileIndex — Represents the “FileIndex” data/behaviour used in this file. - LangPlugin — Represents the “LangPlugin” data/behaviour used in this file.

Functions - _get_cached_parser() — (internal helper) looks up / reads the cached parser step. - _build_cfg_for_function() — (internal helper) builds / writes the code control-flow for function step. - _extract_control_flow_nodes() — (internal helper) reads and cleans up the control flow nodes step. - _py_extract_reflection_facts() — (internal helper) Provides the “py extract dynamic code lookup facts” logic. - _java_extract_reflection_facts() — (internal helper) Provides the “java extract dynamic code lookup facts” logic. - _cs_extract_reflection_facts() — (internal helper) Provides the “cs extract dynamic code lookup facts” logic. - _py_extract_framework_markers() — (internal helper) Provides the “py extract web framework markers” logic. - _java_extract_framework_markers() — (internal helper) Provides the “java extract web framework markers” logic. - _cs_extract_framework_markers() — (internal helper) Provides the “cs extract web framework markers” logic. - _py_extract_response_dataflow() — (internal helper) Provides the “py extract response data movement” logic. - _java_extract_response_dataflow() — (internal helper) Provides the “java extract response data movement” logic. - _cs_extract_response_dataflow() — (internal helper) Provides the “cs extract response data movement” logic. - _py_text() — (internal helper) Provides the “py text” logic. - _py_leftmost_identifier() — (internal helper) Provides the “py leftmost identifier” logic. - _py_snippet() — (internal helper) Provides the “py snippet” logic. - _py_extract() — (internal helper) Provides the “py extract” logic. - _text_of() — (internal helper) Provides the “text of” logic. - _snippet_at() — (internal helper) Provides the “snippet at” logic. - _scope_at() — (internal helper) Provides the “in-scope set at” logic. - _java_leftmost() — (internal helper) Provides the “java leftmost” logic. - _java_extract() — (internal helper) Provides the “java extract” logic. - _js_leftmost() — (internal helper) Provides the “js leftmost” logic. - _js_extract() — (internal helper) Provides the “js extract” logic. - _go_leftmost() — (internal helper) Provides the “go leftmost” logic. - _go_import_spec() — (internal helper) Provides the “go import rule specification” logic. - _go_extract() — (internal helper) Provides the “go extract” logic. - _cs_leftmost() — (internal helper) Provides the “cs leftmost” logic. - _cs_extract() — (internal helper) Provides the “cs extract” logic. - _semantic_sink_override() — (internal helper) Provides the “semantic dangerous operation override” logic. - _match_call() — (internal helper) checks / decides the call value. - _build_spec_index() — (internal helper) builds / writes the rule specification index step. - _py_extract_field_facts() — (internal helper) Provides the “py extract field facts” logic. - _java_extract_field_facts() — (internal helper) Provides the “java extract field facts” logic. - _cs_extract_field_facts() — (internal helper) Provides the “cs extract field facts” logic. - scan_file() — Runs the file value.

pipeline/stages/s0_seed.py

277 LOC. Step 0 — Static seed.

Classes - SeedPackage — Represents the “SeedPackage” data/behaviour used in this file. - __post_init__() — Python built-in method (__post_init__). - __bool__() — Python built-in method (__bool__).

Functions - _filter_step0_languages() — (internal helper) checks / decides the step0 languages step. - run() — Runs the run value. - _resolve_rulepacks() — (internal helper) checks / decides the rulepacks value. - _parse_sarif() — (internal helper) reads and cleans up the machine-readable report value.

pipeline/stages/s11_validate.py

56 LOC. s11_validate — thin wrapper that hooks the validation package into the pipeline.

Functions - run() — Runs the run value.

pipeline/stages/s1_autoexclude.py

368 LOC. Auto-derive a per-target step1 exclusion overlay.

Functions - _survey() — (internal helper) Provides the “survey” logic. - _extract_yaml() — (internal helper) reads and cleans up the config file value. - _norm_list() — (internal helper) reads and cleans up the list value. - run() — Runs the run value.

pipeline/stages/s1_preprocess.py

1548 LOC. Step 1 — Claude CLI explores the repo agentically and emits a ContextPackage.

Functions - _exclusion_sets() — (internal helper) Provides the “skip list sets” logic. - _seed_repo_kinds() — (internal helper) Provides the “seed code repository kinds” logic. - _should_escalate_gap_fill() — (internal helper) checks / decides the widen scope gap fill step. - glob_hit() — Provides the “file pattern hit” logic. - _norm_rel() — (internal helper) reads and cleans up the rel value. - _resolve_scope_path() — (internal helper) checks / decides the in-scope set path step. - _walk_repo() — (internal helper) Provides the “walk code repository” logic. - _flatten_keys() — (internal helper) reads and cleans up the keys value. - _shape_hash() — (internal helper) reads and cleans up the hash value. - _suspicious_set() — (internal helper) Provides the “suspicious set” logic. - _rep_score() — (internal helper) Provides the “rep score” logic. - _dedup_configs() — (internal helper) Provides the “duplicate removal configs” logic. - q_join() — Provides the “q join” logic. - q_split() — Provides the “q split” logic. - q_file() — Provides the “q file” logic. - q_name() — Provides the “q name” logic. - _resolve_py_import_module() — (internal helper) checks / decides the py import module step. - _scan_imports() — (internal helper) runs the imports value. - _file_matches_import() — (internal helper) Provides the “file matches import” logic. - _resolve_callee_files() — (internal helper) checks / decides the callee files step. - _scan_defs() — (internal helper) runs the defs value. - _seed_covered_languages() — (internal helper) Provides the “seed covered languages” logic. - _merge_graph_artifacts() — (internal helper) adds detail to the code relationship map artifacts step. - _supplement_call_graph() — (internal helper) Provides the “supplement call code relationship map” logic. - run() — Runs the run value.

pipeline/stages/s2_threatmodel.py

611 LOC. Step 2 — Application threat model.

Functions - _repo_kind() — (internal helper) Provides the “code repository kind” logic. - _baseline_block() — (internal helper) Provides the “baseline block” logic. - _read_capped() — (internal helper) looks up / reads the capped value. - _gather_evidence() — (internal helper) looks up / reads the evidence value. - _build_user_prompt() — (internal helper) builds / writes the user prompt step. - run() — Runs the run value.

pipeline/stages/s3_decompose.py

1521 LOC. Step 3 — the strategist LLM receives the ContextPackage (no raw code) and produces a

Functions - run() — Runs the run value. - _normalize_chunk_files() — (internal helper) reads and cleans up the analysis unit files step. - _is_source() — (internal helper) checks / decides the untrusted input point value. - _file_call_graph() — (internal helper) Provides the “file call code relationship map” logic. - _cohesive_groups() — (internal helper) Provides the “cohesive groups” logic. - _report_threat_coverage() — (internal helper) builds / writes the attack scenario coverage step. - _tok() — (internal helper) Provides the “token/cost estimate” logic. - _is_config_file() — (internal helper) checks / decides the config file step. - _threat_text() — (internal helper) Provides the “attack scenario text” logic. - _matches_threat_surface() — (internal helper) checks / decides the attack scenario surface step. - _candidate_files_for_threat() — (internal helper) looks up / reads the files for attack scenario step. - _add_threat_surface_fallback_chunks() — (internal helper) adds detail to the attack scenario surface fallback analysis units step. - _pick_hop_files() — (internal helper) checks / decides the hop files step. - _qnode_for_file_line() — (internal helper) Provides the “a node in the code graph for file line” logic. - _sink_qnodes_for_sink() — (internal helper) Provides the “dangerous operation nodes in the code graph for dangerous operation” logic. - _seed_paths_for_entry() — (internal helper) Provides the “seed paths for entry” logic. - _add_taint_chunks() — (internal helper) adds detail to the untrusted-data flow analysis units step. - _bfs_to_sinks() — (internal helper) Provides the “bfs to dangerous operations” logic. - _size_for() — (internal helper) Provides the “size for” logic. - _char_budget() — (internal helper) Provides the “char spending limit” logic. - _split_oversize_risk_chunks() — (internal helper) reads and cleans up the oversize risk analysis units step. - _pack() — (internal helper) Provides the “pack” logic. - _catchall_eligible() — (internal helper) Provides the “catchall eligible” logic. - _lang_of_file() — (internal helper) Provides the “lang of file” logic. - _reachable_files() — (internal helper) Provides the “reachable files” logic. - _reachable_only_too_sparse() — (internal helper) Provides the “reachable only too sparse” logic. - _add_catchall_chunks() — (internal helper) adds detail to the catchall analysis units step. - _add_specialist_chunks() — (internal helper) adds detail to the specialist analysis units step. - _has_batch_surface() — (internal helper) checks / decides the batch surface step. - _scan_any() — (internal helper) runs the any value. - _has_authz_surface() — (internal helper) checks / decides the authz surface step. - _gate_specialists() — (internal helper) checks / decides the specialists value. - _mk_specialist() — (internal helper) Provides the “mk specialist” logic. - _specialist_focus_entry_points() — (internal helper) Provides the “specialist focus entry points” logic. - _mk_catchall() — (internal helper) Provides the “mk catchall” logic. - _report_chunk_loc() — (internal helper) builds / writes the analysis unit lines of code step. - _count_chars() — (internal helper) Provides the “count chars” logic. - _count_loc() — (internal helper) Provides the “count lines of code” logic.

pipeline/stages/s4_deepdive.py

1518 LOC. Step 4 — Deep-dive each chunk N times, then majority-vote.

Functions - build_research_lens() — Builds / writes the research lens step. - build_system_prompt() — Builds / writes the system prompt step. - _effective_runs() — (internal helper) Provides the “effective runs” logic. - run() — Runs the run value. - _deepdive_chunk() — (internal helper) Provides the “deepdive analysis unit” logic. - _collapse_across_chunks() — (internal helper) Provides the “collapse across analysis units” logic. - _single_run() — (internal helper) Provides the “single run” logic. - _trust_context_block() — (internal helper) Provides the “trust context block” logic. - _build_prompt() — (internal helper) builds / writes the prompt value. - _norm_rel_path() — (internal helper) reads and cleans up the rel path step. - _parse_ref_file_line() — (internal helper) reads and cleans up the ref file line step. - _short_symbol() — (internal helper) Provides the “short symbol” logic. - _compact_taint_evidence_block() — (internal helper) Provides the “compact untrusted-data flow evidence block” logic. - _build_confirm_refute_prompt() — (internal helper) builds / writes the confirm refute prompt step. - _mask_pan() — (internal helper) hides sensitive data in the pan value. - _redact_source() — (internal helper) hides sensitive data in the untrusted input point value. - _load_chunk_code() — (internal helper) looks up / reads the analysis unit code step. - _slice_mode() — (internal helper) Provides the “slice mode” logic. - _load_graph_slice() — (internal helper) looks up / reads the code relationship map slice step. - _load_taint_slice() — (internal helper) looks up / reads the untrusted-data flow slice step. - _merge_ranges() — (internal helper) adds detail to the ranges value. - _count_lines() — (internal helper) Provides the “count lines” logic. - _neighbor_context() — (internal helper) Provides the “neighbor context” logic. - _excerpt() — (internal helper) Provides the “excerpt” logic. - _load_files_full() — (internal helper) looks up / reads the files full step. - _load_sliding_window() — (internal helper) looks up / reads the sliding window step. - _windows_for_entrypoints() — (internal helper) Provides the “windows for entrypoints” logic. - _windows_for_anchors() — (internal helper) Provides the “windows for anchors” logic. - _graph_anchor_lines_for_file() — (internal helper) Provides the “code relationship map anchor lines for file” logic.

pipeline/stages/s5_prefilter.py

249 LOC. Step 5 — Deterministic pre-filter.

Functions - _is_secret_class() — (internal helper) checks / decides the secret class step. - _gate() — (internal helper) checks / decides the rule check value. - _ast_backfill_evidence() — (internal helper) Provides the “ast backfill evidence” logic. - run() — Runs the run value.

pipeline/stages/s6_verify.py

426 LOC. Step 6 — Adversarial verification.

Functions - run() — Runs the run value. - _verify_one() — (internal helper) checks / decides the one value. - _controls_for() — (internal helper) Provides the “controls for” logic. - _callers_of() — (internal helper) Provides the “callers of” logic. - _candidate_qnodes_for_finding() — (internal helper) looks up / reads the nodes in the code graph for candidate issue step. - _callgraph_context_for_finding() — (internal helper) Provides the “map of which code calls what context for candidate issue” logic. - _build_user_prompt() — (internal helper) builds / writes the user prompt step. - _parse_verdict() — (internal helper) reads and cleans up the pass/fail decision value. - _drop() — (internal helper) Provides the “drop” logic.

pipeline/stages/s7_dedup.py

371 LOC. Step 7 — Deduplication.

Functions - _make_dup_location() — (internal helper) builds / writes the dup location step. - _dedup_locations() — (internal helper) Provides the “duplicate removal locations” logic. - _attach_duplicates() — (internal helper) adds detail to the duplicates value. - _collapse_trivial() — (internal helper) Provides the “collapse trivial” logic. - prefilter() — Provides the “early filter” logic. - run() — Runs the run value. - _semantic_dedup() — (internal helper) Provides the “semantic duplicate removal” logic. - _parse_dedup_output() — (internal helper) reads and cleans up the duplicate removal output step. - _graph_context_for_finding() — (internal helper) Provides the “code relationship map context for candidate issue” logic.

pipeline/stages/s8_chain.py

519 LOC. Step 8 — the chain-analysis LLM sees ALL findings together and:

Functions - _reachability_for_finding() — (internal helper) Provides the “reachability for candidate issue” logic. - run() — Runs the run value. - _coerce_list_payload() — (internal helper) reads and cleans up the list payload step. - _unranked_report() — (internal helper) Provides the “unranked report” logic. - _hydrate_report() — (internal helper) reads and cleans up the report value. - _build_prompt() — (internal helper) builds / writes the prompt value. - _sev_from_band() — (internal helper) Provides the “severity from band” logic. - _final_severity() — (internal helper) Provides the “final severity” logic. - _offensive_rank() — (internal helper) Provides the “offensive rank” logic. - _severity_sort_order() — (internal helper) Provides the “severity sort order” logic. - _coerce_sev() — (internal helper) reads and cleans up the severity value.


See also: HLD · LLD · Data Flow · module index.