Trailmark

Trailmark

Parses source code into a directed graph of functions, classes, calls, and semantic metadata for security analysis.

When to Use

• Mapping call paths from user input to sensitive functions
• Finding complexity hotspots for audit prioritization
• Identifying attack surface and entrypoints
• Understanding call relationships in unfamiliar codebases
• Security review or audit preparation across polyglot projects
• Adding LLM-inferred annotations (assumptions, preconditions) to code units
• Importing external binary-analysis graphs to connect source and binary views
• Querying transitive slices, entrypoint paths, subgraph edges, or type references
• Producing graph evidence for one suspicious function or candidate finding
• Pre-analysis before mutation testing (genotoxic skill) or diagramming

When NOT to Use

• Single-file scripts where call graph adds no value (read the file directly)
• Architecture diagrams not derived from code (use the diagramming-code skill or draw by hand)
• Mutation testing triage (use the genotoxic skill, which calls trailmark internally)
• Runtime behavior analysis (trailmark is static, not dynamic)

Rationalizations to Reject

Installation

MANDATORY: If trailmark is not found, install the CLI before doing anything else:

uv tool install trailmark

A tool install provides the CLI only — it does not make import trailmark resolvable. Run the Python snippets in this skill with uv run --with trailmark python -; that, not installation, is the fix for an import error or ModuleNotFoundError in a snippet.

DO NOT fall back to "manual verification", "manual analysis", or reading source files by hand as a substitute for running trailmark. The tool must be installed and used programmatically. If installation fails, report the error to the user instead of silently switching to manual code reading.

Version Gate

Trailmark 0.4.0 expands the graph model and query surface, and 0.5.0 adds a SQL parser, repository-link configuration, and richer entrypoint metadata. Before using a feature listed as v0.4+ or v0.5+, check the installed version:

trailmark --version 2>/dev/null || uv run trailmark --version 2>/dev/null

Compare the reported version numerically (not lexically). 0.4.0 or newer means the full v0.4 surface is available. The version command itself was added in 0.2.2, so a failure means either a pre-0.2.2 install or trailmark missing entirely — distinguish with trailmark analyze --help. When working programmatically, probe with hasattr() and fall back instead of assuming a v0.4-only method exists:

if hasattr(engine, "subgraph_edges"):    edges = engine.subgraph_edges("tainted")else:    # v0.2 fallback: filter engine.to_json() edges whose endpoints    # are both in engine.subgraph("tainted")    edges = []

v0.2-safe baseline: CLI analyze, diff, entrypoints, augment, and --language auto; QueryEngine.from_directory(), callers_of(), callees_of(), paths_between(), ancestors_of(), reachable_from(), entrypoint_paths_to(), complexity_hotspots(), attack_surface(), summary(), to_json(), preanalysis(), annotate(), annotations_of(), nodes_with_annotation(), clear_annotations(), findings(), subgraph(), subgraph_names(), diff_against(), augment_sarif(), and augment_weaudit().

Added in 0.2.2: CLI --version flag and version subcommand.

Added in 0.3.x: the trailmark.parse module with module-level detect_languages() and supported_languages(). detect_languages() itself is v0.2-safe via from trailmark.query.api import detect_languages (kept as a deprecated alias in 0.3+); supported_languages() has no 0.2.x equivalent.

v0.4+ features: native diagram subcommand; expanded parser coverage; proxy nodes for unresolved calls; node origins; binary graph augmentation via augment_binary(); connect_subgraphs(); subgraph_edges(); generic_parameters(); and type_references().

v0.5+ features: sql parser (PostgreSQL-oriented schemas, tables, views, functions, procedures, dependencies); node kinds schema, table, view, procedure; .trailmark/links.toml repository-link configuration (see Repository Links below), including proxy.external:<symbol> nodes for declared external endpoints; repository links, unresolved-call proxies, and type_uses edges now materialize for single-language directory parses (0.4 emitted them only for polyglot parses); Solidity entrypoints detected from parser metadata (interfaces excluded; solidity_visibility, solidity_mutability, solidity_override, solidity_container_kind, and solidity_overridden_by node attributes); attack_surface() entries carry an attributes key when the node has attributes; TypeScript resolves receivers assigned with new ConcreteClass(); C# file-scoped namespaces.

v0.5.0 adds no new QueryEngine methods, so hasattr(engine, ...) cannot detect it. Gate v0.5 features on the reported version, or probe structurally:

from trailmark.models.nodes import NodeKind
has_v05 = "SCHEMA" in NodeKind.__members__  # sql kinds are 0.5+

Quick Start

# Auto-detect and merge every supported language under the treeuv run trailmark analyze --language auto --summary {targetDir}
# Explicit languages (single language or comma-separated list)uv run trailmark analyze --language rust {targetDir}uv run trailmark analyze --language python,rust {targetDir}
# Complexity hotspotsuv run trailmark analyze --language auto --complexity 10 {targetDir}
# Entrypoint inventory and structural diff (v0.2-safe)uv run trailmark entrypoints --language auto {targetDir}uv run trailmark diff --language auto --repo {repoDir} main HEAD --json
# Version report (0.2.2+)uv run trailmark --version
# v0.4+: native diagram commanduv run trailmark diagram -t {targetDir} -T call-graph -f main --depth 2

Programmatic API

# trailmark.parse is a 0.3+ module; on 0.2.x import detect_languages from# trailmark.query.api instead (supported_languages has no 0.2.x equivalent)from trailmark.parse import detect_languages, supported_languagesfrom trailmark.query.api import QueryEngine
# Ask the installed Trailmark build what it supportssupported_languages()detect_languages("{targetDir}")
# Prefer auto for unknown or polyglot trees; use explicit lists when neededengine = QueryEngine.from_directory("{targetDir}", language="auto")engine = QueryEngine.from_directory("{targetDir}", language="python,rust")
engine.callers_of("function_name")engine.callees_of("function_name")engine.paths_between("entry_func", "db_query")engine.complexity_hotspots(threshold=10)engine.attack_surface()engine.summary()engine.to_json()
# Transitive slices and entrypoint path queries (v0.2-safe)engine.ancestors_of("sensitive_sink")engine.reachable_from("entry_func")engine.entrypoint_paths_to("sensitive_sink")
# v0.4+: connect named subgraphsif hasattr(engine, "connect_subgraphs"):    engine.connect_subgraphs("tainted", "privilege_boundary")
# Run pre-analysis (blast radius, entrypoints, privilege# boundaries, taint propagation)result = engine.preanalysis()
# Query subgraphs created by pre-analysisengine.subgraph_names()engine.subgraph("tainted")engine.subgraph("high_blast_radius")engine.subgraph("privilege_boundary")engine.subgraph("entrypoint_reachable")if hasattr(engine, "subgraph_edges"):    engine.subgraph_edges("tainted")
# Add LLM-inferred annotationsfrom trailmark.models import AnnotationKind
engine.annotate("function_name", AnnotationKind.ASSUMPTION,                "input is URL-encoded", source="llm")
# Query annotations (including pre-analysis results)engine.annotations_of("function_name")engine.annotations_of("function_name",                       kind=AnnotationKind.BLAST_RADIUS)engine.annotations_of("function_name",                       kind=AnnotationKind.TAINT_PROPAGATION)engine.nodes_with_annotation(AnnotationKind.FINDING)engine.clear_annotations("function_name", kind=AnnotationKind.ASSUMPTION)
# v0.4+: generic/type-reference and binary augmentation APIsif hasattr(engine, "generic_parameters"):    engine.generic_parameters("GenericTypeOrFunction")if hasattr(engine, "type_references"):    engine.type_references("function_name")if hasattr(engine, "augment_binary"):    engine.augment_binary("binary_graph.json")

Pre-Analysis Passes

Always run engine.preanalysis() before handing off to genotoxic or diagramming-code skills. Pre-analysis enriches the graph with four passes:

1. Blast radius estimation — counts downstream and upstream nodes per function, identifies critical high-complexity descendants
2. Entry point enumeration — maps entrypoints by trust level, computes reachable node sets
3. Privilege boundary detection — finds call edges where trust levels change (untrusted -> trusted)
4. Taint propagation — marks all nodes reachable from untrusted entrypoints

Results are stored as annotations and named subgraphs on the graph.

For detailed documentation, see references/preanalysis-passes.md [blocked].

Language Selection

Do not hardcode a stale language table in downstream workflows. Ask the installed Trailmark build what it supports:

from trailmark.parse import detect_languages, supported_languages
supported_languages()detect_languages("{targetDir}")

CLI patterns:

# Auto-detect and mergeuv run trailmark analyze --language auto {targetDir}
# Explicit list for a known polyglot targetuv run trailmark analyze --language python,rust {targetDir}

As of Trailmark 0.5.0, parser names include: python, javascript, typescript, php, ruby, c, cpp, c_sharp, java, go, rust, solidity, cairo, circom, haskell, erlang, masm, swift, objc, kotlin, dart, move, tact, func, sway, rego, proto, thrift, graphql, and sql (added in 0.5.0; PostgreSQL-oriented, .sql files). Treat this list as documentation, not a source of truth; call supported_languages() on the installed build before relying on a parser.

Repository Links (v0.5+)

Parsers cannot see cross-language calls (FFI, RPC, IPC, contract invocation) or edges into external systems. Declare them in .trailmark/links.toml at the analysis root and Trailmark materializes the edges on every parse — this is a stable public configuration interface:

[[link]]source = "backend:submit"target = "contract:Verifier.verify"kind = "calls"                 # any EdgeKind; defaults to callsconfidence = "certain"         # certain | inferred | uncertain; defaults to inferreddescription = "JSON-RPC eth_call"
[[link]]source = "backend:notify"target = "payments-webhook"target_external = true         # required because target is unresolved

Endpoint references may be exact node IDs or unique names/suffixes. Validation fails closed: ambiguous references, unknown internal endpoints, invalid enum values, and malformed TOML raise ValueError rather than silently weakening the graph. source_external = true / target_external = true permit an unresolved endpoint by creating a proxy.external:<symbol> node. Configured edges carry a configured_by = .trailmark/links.toml attribute so they are distinguishable from parser-derived edges.

Use this when the audit spans an FFI/RPC boundary the rationalization table warns about: declare the boundary edges first, then path and taint queries cross them like any other call edge.

Graph Model

Node kinds: function, method, class, module, struct, interface, trait, enum, namespace, contract, library, template; v0.4+ also materializes unresolved references as proxy nodes; v0.5+ adds schema, table, view, and procedure for SQL graphs.

Node origins: v0.4+ nodes may carry origin source, proxy, binary, or synthetic. v0.2 exports may omit origin.

Edge kinds: calls, inherits, implements, contains, imports; v0.4+ adds resolves_to, type_uses, specializes, and corresponds_to.

Edge confidence: certain (direct call, self.method()), inferred (attribute access on non-self object), uncertain (dynamic dispatch)

Per Code Unit

• Parameters with types, return types, exception types
• Cyclomatic complexity and branch metadata
• Docstrings
• Annotations: assumption, precondition, postcondition, invariant, blast_radius, privilege_boundary, taint_propagation, finding, audit_note (last two set by augment_sarif / augment_weaudit)

Per Edge

• Source/target node IDs, edge kind, confidence level

Project Level

• Dependencies (imported packages)
• Entrypoints with trust levels and asset values
• Named subgraphs (populated by pre-analysis)

Key Concepts

Declared contract vs. effective input domain: Trailmark separates what a function declares it accepts from what can actually reach it via call paths. Mismatches are where vulnerabilities hide:

• Widening: Unconstrained data reaches a function that assumes validation
• Safe by coincidence: No validation, but only safe callers exist today

Edge confidence: Dynamic dispatch produces uncertain edges. Account for confidence when making security claims.

Proxy nodes (v0.4+): Unresolved calls are preserved as nodes such as proxy.unresolved:<symbol>. Do not treat these as source code functions; use them to identify resolution gaps, dynamic dispatch, external APIs, or binary linkage candidates. v0.5+ also emits proxy.external:<symbol> nodes for endpoints declared external in .trailmark/links.toml.

Reachability is not taint: entrypoint_paths_to() and the taint subgraph answer different questions. Path queries report call-graph reachability; preanalysis taint marks nodes reachable from untrusted entrypoints as a coarse signal. Trailmark does not perform interprocedural taint analysis — do not present either as proof that attacker-controlled data reaches a sink.

Binary augmentation (v0.4+): engine.augment_binary() imports an external binary-analysis graph JSON file. Trailmark connects it to source nodes when possible; it does not disassemble binaries itself.

Subgraphs: Named collections of node IDs produced by pre-analysis. Query with engine.subgraph("name"). Available after engine.preanalysis().

Query Patterns

See references/query-patterns.md [blocked] for common security analysis patterns.

See references/preanalysis-passes.md [blocked] for pre-analysis pass documentation.

Use trailmark-finding-triage when the user has one concrete candidate finding, SARIF result, weAudit annotation, suspicious function, or report excerpt and needs a handoff-ready reachability and blast-radius evidence packet.

Use trailmark-variant-neighborhood after one seed issue is known and the user needs graph-derived variant candidates for variant-analysis, Semgrep, CodeQL, or manual review.