Crate agentstatedeveloper

Expand description

AgentStateDeveloper: code-level context and an audit overlay for agent-authored code.

This is the library entry point. It re-exports everything from agentstatedeveloper-core (the engine, schema, stores and index pipeline) and default_adapters, which returns every built-in language adapter.

let adapters = agentstatedeveloper::default_adapters();
assert!(!adapters.is_empty());

The asd command-line tool is in agentstatedeveloper-cli, and the MCP server is in agentstatedeveloper-mcp.

Modules§

adapter
audit: Audit-log event stream — OSS data format.
brief: Plan D t-001 + t-007: shared brief-output projections.
calibration: Plan J t-015: confidence-bucket calibration helpers.
candidates: Shared candidate-selection logic used by CLI commands and the MCP server.
community: Community detection over the call graph (Plan competitive-harvest t-009).
conclusions_export: Plan B t-004: shared conclusion-export pipeline used by both the CLI (asd conclusions export) and MCP (conclusions_export tool).
context: Per-symbol context assembly for asd context-for (CLI) and the mcp__asd__context_for MCP tool.
cross_service: Cross-service / cross-process edges (Plan competitive-harvest t-002).
dataflow: Intra-process data-flow edges (Plan competitive-harvest t-002, slice 4).
doc_adapters: Document adapters: produce SearchDoc chunks from non-code files.
edge_confidence: Confidence for call-graph edges (Plan competitive-harvest t-013).
effects
engine
error
feedback: Feedback store — durable (query, symbol, verdict) triples.
gc: ASD’s garbage-collection retention policy, shared by asd gc and asd-serve’s sweep preview so the two can never disagree about what a sweep keeps.
help: On-demand instruction disclosure (help) — the asd half of the cross-binary help system shared with ctx (CTXone).
index
index_pipeline: Shared indexing pipeline used by both the CLI asd index command and the MCP reindex tool.
ledger
ledger_integrity: Ledger integrity across the three places an entry lives.
map: asd map core — initial-read project summary.
overview: Read-only “orient me” reports shared by the CLI and the asd-mcp server.
paths: ASG path convention for ASD data.
policy: Policy gate.
prepare_change: Plan E t-006: shared helpers for the prepare-change pipeline.
rebind: Moving a symbol’s ledger history onto another symbol.
recipes: Plan C t-004: change-intent recipes.
registry: Shared ASD repo registry — ~/.config/asd/repos.toml.
repair: ASG state repair — scan for corruption and apply safe auto-corrections.
schema
scorecard: The five-dimension benchmark scorecard — truth, feedback, change, uncertainty, workflow — plus the data-quality and token-economy blocks that qualify them.
scratch: Scratchpad store for ephemeral agent working notes.
search_fts: FTS5-backed search index for ASD symbols.
ser_helpers: Serde skip-predicates for agent-mode token economy.
sidecar: On-disk sidecar for the “git roundtrip” promise.
sidecar_config: Plan K t-007: .asd/config.toml reader for opt-in sidecar sharding. Today a single shape: per-package sharding via [sidecar] shard_by = "package". Default is “class” (one file per conclusion class, current behavior).
symbol
test_summary: Compact, failures-only summaries of test-runner output (Plan competitive-harvest t-005). Reads a runner’s verbose output and keeps only what an agent needs to act: the pass/fail counts and each failure with a little context — typically a ~90% token cut versus the raw log.
thinking: Plan G t-006: read-side helper for surfacing captured thinking (Hypothesis / MentalModel / FailedAttempt / OpenQuestion) into prepare-change and context-for responses.
transitive: Transitive effect propagation.
trust: State Trust Score — machine-readable rollup of ASD semantic state quality.
workflow: Workflow Recipe Detection + Task Close Evidence Quality.

Structs§

AnnotatedOwner: An owner with the evidence source that identified them.
ApprovalOutcome: Outcome of LedgerStore::approve_entry. Carries the updated entry and whether it was already approved (caller idempotency hint).
AsgEffectStore
AsgFeedbackStore
AsgIndexStore: ASD index reader/writer.
AsgLedgerStore
AsgScratchStore: ScratchStore backed by an ASG Repository.
AuditEvent: One audit entry. Flat, JSON-serializable. Shared data format — event_hash / prev_event_hash are empty strings / None when the OSS binary emits (via NullSink) and filled in by the Pro sink.
Author
CacheWarmSummary: Result of Engine::warm_caches.
CallEdge: A directed call edge from one qname to another, produced by a language adapter’s intra-module call-graph extractor.
CandidateAggregates: Holds the six parallel collections built up by the per-candidate scan plus the top symbol id (driver for downstream BFS) and the seen_inv dedup set (threaded into the subsequent caller-invariant propagation pass so both invariant sources share one namespace).
CleanFilter: Predicate used by ScratchStore::clean_entries to select entries for permanent deletion.
CollectResult: Result of collecting source files under a path.
CoveringTest: A test symbol that likely covers a given impl symbol.
CrossServiceEdge: A matched cross-service edge: an outbound endpoint linked to an inbound endpoint sharing a contract key. Computed on demand, not materialized.
DataFlowEdge: A resolved data-flow edge: a value flows from a call site in from into param of the callee to.
DataQuality: Distinguishes intentional clean-room state from unexpected data loss.
DetectedDataFlow: A data-flow site detected by an adapter, before the pipeline resolves the callee’s parameter name and symbol identity. The adapter knows the callee by qname and the argument’s position; the pipeline maps position → param name using the callee’s signature.
DetectedEndpoint: An endpoint detected by a language adapter, before the index pipeline enriches it with this repo’s repo_id and the owning symbol’s symbol_id. Adapters work in qnames, so a DetectedEndpoint names its owner by qname.
DocHit: A ranked document search hit.
DynamicDispatchHint: Plan L t-005: one dynamic-dispatch call site detected by an adapter. pattern is a short label (getattr / callback-arg / metaclass); snippet is the source line trimmed for display in CLI warnings.
EdgeEvidence: Accumulated evidence for one directed call edge.
Effect
EffectDecl
EndpointRef: A lightweight reference to one end of a cross-service edge.
Engine: The top-level ASD engine. Owns an ASG repository, a policy gate, an audit sink, and optional commercial ratify operations. Cheap to construct; shared across CLI, MCP, and library consumers.
EvidenceQuality: Per-field evidence quality breakdown for the task closure.
FeedbackEntry: A single feedback record: a (query, symbol, verdict) triple stored durably.
FeedbackImpact: Per-result explanation of which feedback verdict affected a search hit.
FeedbackMetrics: Apply feedback verdicts as score adjustments after candidate selection.
FeedbackState: Explicit description of feedback availability for the current query.
FilePolicyGate
FileRecency: Per-file recency metadata derived from a single git log call.
FtsFilters: Filters applied before BM25 ranking.
FtsHit: A single ranked result from FTS search.
HydrateSummary: Result of hydrate_from_dir. missing_schema_version is true when the sidecar exists but has no meta/schema-version file; hydrate still proceeds but callers should surface the mismatch.
IndexSummary: Summary returned by run_index.
LedgerCounts: Ledger entry counts in each store, plus the gaps between them.
LedgerEntry
LedgerRestoreReport: Outcome of restore_missing_ledger_entries.
Mismatch
NullSink: Swallows every event. The OSS default — tamper-evident file-backed audit is a commercial feature (see agentstatedeveloper-audit-pro).
OwnershipSignal: Ownership signal returned by discover_symbol_ownership.
ParsedSymbol: Parsed symbol returned by a language adapter. Carries enough info to build a full Symbol node.
PermissivePolicyGate
PolicyFile
PolicyRule
Position
Rebind: Record that a symbol was renamed/moved so its ledger history follows it. Written to /asd/v1/rebinds/<from_symbol_id> at rename time.
RebindOutcome: What one rebind moved.
RecoverySuggestion: Recovery quality estimate for a single scoped suggestion.
RepairIssue: A single integrity issue detected in the ASG.
RepairReport: Result returned by repair_asg.
ResolvedSymbol: Full symbol resolution result from the FTS table.
ReviewOutcome: Outcome of LedgerStore::reject_entry / withdraw_entry. Same idempotency shape as ApprovalOutcome.
RuntimeEvidence: Accumulated runtime-trace evidence for a symbol’s declared effects.
ScratchEntry: Ephemeral working note scoped to a symbol and/or named workflow.
ScratchFilter: Predicate applied by ScratchStore::list_entries. All Some fields must match (AND semantics).
SearchDoc: A searchable chunk from a non-code file (markdown, config, manifest, etc.).
SearchDocsDb: FTS index for document/resource chunks (separate from symbol FTS).
SearchFtsDb
ServiceEndpoint: A detected service-boundary point owned by a local symbol.
ServiceManifest: A repo’s exported endpoints — the unit shared for cross-repo matching.
Situation: Context passed to every policy evaluation call. Keys are populated by the call site so selectors can key on symbol_id, agent_id, etc.
StaleWarning: AcmeFlow refinement (1.0.77): same as stale_warning but returns a struct so callers can distinguish a soft “index is old but results came back fine” hint from a hard “FTS is broken / empty” alert. The MCP response includes stale_severity as a machine-readable field so downstream UIs can render appropriately (Craig decision Q7: yes, future-proof rendering).
Symbol
SymbolMeta: Lightweight symbol metadata stored in the asd_symbols_meta SQLite table.
SyncSummary: Result of sync_to_dir. Counts what was written; the schema version is always stamped.
TransitiveEffect
TrustScore: Rolled-up state trust score for the current ASD workspace.
TrustSignals: Raw signal values gathered before scoring. Exposed for debugging and for truth-drift history snapshots.
UncertaintyReport: Machine-readable uncertainty rollup.
UnresolvedCall: Plan L t-006: one call site the static walker tried to resolve but couldn’t. May be stdlib (print, len), third-party (an unindexed dependency), or genuinely dynamic. Aggregated by the indexer to expose the static-resolution coverage gap so agents don’t assume missing edges are bugs.
Verification
WorkflowSummary: Rolled-up workflow summary emitted in task-close JSON and persisted to .asd/workflow-sessions.jsonl.
WorkspaceSymbols: Workspace-wide context passed to adapters when resolving call edges across module boundaries.

Enums§

AsdError
AuthorKind
ConclusionClass: The conclusion-class buckets that drive .asd/conclusions/*.jsonl export and the Plan B sidecar redesign. Each class corresponds to one JSONL file; LedgerKind variants are bucketed via conclusion_class().
DataFlowKind: How a value reaches its sink. Currently only argument→parameter; field-access chains are a planned addition.
Decision
Direction: Whether an endpoint exposes/handles a contract (inbound) or consumes/calls it (outbound).
DocKind: Broad category for a document chunk.
EffectCategory: Effect category — either a well-known built-in or a user-defined domain string (e.g. "midi.send", "audio.graph.connect").
Evidence
FeedbackVerdict: Verdict an agent or user assigns to a search result.
IssueSeverity: Severity level for a RepairIssue.
LedgerKind
OwnerSignalSource: The evidence source that identified an owner.
RoleTag: Plan C t-002: first-class role-tag vocabulary. The optional LedgerEntry.role field is free-form String (so unknown tags don’t break old data), but CLI / MCP / API layers validate against this enum at write time and warn on unknown values.
ScratchStatus: Status of a ScratchEntry. Transitions: Draft → Promoted or Discarded.
SidecarState: Observable lifecycle state of the on-disk sidecar.
StaleSeverity
SymbolKind
Transport: Transport a service endpoint communicates over.
UncertaintyReason: A single, machine-readable uncertainty signal.
VerificationSource
VerificationStatus

Constants§

AGENT_DEFAULT_BUDGET: Default token budget for --agent mode (8 000 tokens ≈ 32 000 chars).
ASD_PATH_PREFIX
ASD_SCHEMA_VERSION
CLIFF_RATIO_THRESHOLD: Cliff detection threshold (1.0.85, AcmeFlow refinement #2 deep half): when a consecutive score ratio drops below this, the lower score is treated as the start of a new “also-ran” cohort and the floor is raised to exclude it.
DEFAULT_FEEDBACK_HALF_LIFE_DAYS: Default half-life for feedback boost decay, in days.
FILE_SCORE_FLOOR_RATIO: Relative score floor for file inclusion in likely_edit_files.
HASH_PREFIX: Prefix for blake3-based event hashes. Versioned so we can change the algorithm later without ambiguity. Shared between OSS readers and Pro writers.
SOFT_STALE_THRESHOLD_SECS: AcmeFlow refinement: 24h soft-threshold for handlers that have a “did the query resolve” signal (prepare-change, context_for). Matches a typical dev day — index built this morning is fine all afternoon. Other commands (status, search, investigate) keep the 1h legacy threshold via stale_warning since they ARE the index-health signal.

Traits§

AuditSink
EffectStore
FeedbackStore
IndexStore
LanguageAdapter: Language-specific parsing + effect inference. Implementations live in sibling crates: agentstatedeveloper-python, -typescript, etc.
LedgerStore
PolicyGate
RatifyOps: The three ledger review operations. Implemented by RatifyOpsImpl in agentstatedeveloper-ratify (Team tier). Methods take an explicit repo: &Repository so the impl can be a zero-sized struct — no lifetime or clone needed when stored as Arc<dyn RatifyOps> in Engine.
ScratchStore: Scratch-entry CRUD operations.

Functions§

adapt_document: Detect the kind for a file path and, if recognised, parse it into chunks. Returns None when the file is not handled by any adapter.
aggregate_candidate_data: Main candidate-scan pass. Iterates the ranked candidate list once, populating all six parallel accumulators:
append_workflow_session: Append a compact workflow session record to .asd/workflow-sessions.jsonl. Silently skips on any I/O error so task-close never fails due to logging. Caps the file at 500 lines (drops the oldest when over limit).
apply_feedback_adjustments
apply_file_scope_feedback: Apply file-scoped feedback verdicts to scored.
architecture_overview: One-call “orient me” overview: languages, packages, layers, call-graph communities (clusters), inbound/outbound routes, and hotspots. This is the JSON body asd architecture --agent prints.
assemble_symbol_context: Assemble the per-symbol context object — the building block both CLI asd context-for and MCP context_for emit per requested qname.
build_feedback_state: Convenience wrapper that fetches feedback entries before calling build_feedback_state_from_entries. Use the _from_entries variant when you already hold a list_all() result.
build_feedback_state_from_entries: Build a FeedbackState for query from the feedback store.
build_id_map: id → Symbol map for the current ref. Fast path via the SQLite symbol cache, authoritative git fallback. Shared by the many commands that need to resolve call-graph node ids back to symbols.
canonical_symbol_id: Canonical stable id for a symbol. Survives content edits and path moves; changes only on rename (which triggers a rebind record).
classify_file_role: Plan J t-003: classify a file by its functional ROLE in the project — view / viewmodel / test / example / fixture / script / generated / reference / impl. Distinct from classify_layer, which buckets by architectural tier (presentation / domain / persistence / etc.).
classify_layer: Classify a source file into a workflow layer for asd investigate grouping.
classify_layer_sym: Like classify_layer but uses the symbol’s qualified name as a secondary hint when the file path alone yields "other". Catches common cases where a class named FooViewModel lives in a file named after the app rather than the ViewModel.
cliff_cutoff_index: Compute the precision floor for file scoring. Two-pass:
collect_source_files
compute_index_consistency: Plan J t-005: reconcile the two divergent symbol counts that asd status and asd health were each reporting in isolation.
compute_trust_score: Compute a TrustScore for the ASD workspace at db_path.
compute_uncertainty: Build a structured UncertaintyReport from the signals already computed by the search/investigate/prepare-change pipeline.
confidence_reason: One-line explanation of why a result has high or low confidence.
confidence_scores: Normalize raw scores to [0.1, 1.0] confidence within the result set. The highest-scoring result gets 1.0; the lowest gets 0.1.
contract_hash: Stable 16-hex-char hash of a contract key, for use as a filesystem-/path-safe segment (raw contract keys contain spaces, /, and :).
dead_code_report: Functions/methods with no inbound call edges — a candidate list. Excludes HTTP route handlers, test functions (unless include_tests), and main/dunder methods. This is the JSON body asd dead-code --agent prints.
decay_factor: Pure decay function. Returns a multiplier in [0.0, 1.0] for an observation of age_days against half_life_days. Uses the classical exponential decay: factor = 0.5 ^ (age / half_life).
decay_for_entry: Compute the decay multiplier for an entry whose created_at is known, given the current time. Used at feedback-boost application time to scale Useful verdicts by age.
declared_effect_blast_radius: Per-declarer transitive blast radius, derived entirely from the stored transitive data written by propagate_transitive — no call-graph reachability is recomputed here.
default_adapters: Return one instance of every built-in language adapter.
derive_cold_hints: Heuristically propose a parallel test file path for a given source file.
detect_ambiguous_tokens: Detect query tokens that match too many unrelated files (broad/ambiguous terms).
detect_communities: Detect communities over an undirected graph. nodes should be passed in a stable (e.g. sorted) order for reproducibility. edges are unordered pairs; self-loops and edges to unknown nodes are ignored, duplicates collapse. Returns a dense node → community_id map.
detect_confidence_warnings: Distinguish whether low result confidence comes from ambiguous terms or a sparse index. Returns a vec of typed warning objects with kind + message.
detect_orphaned_entries: Walk every ledger entry and tag those whose symbol_id is no longer present in the qname index as "orphaned" and "orphaned-at:<timestamp>"; clear those tags from entries whose symbol is back.
detect_possible_misses: Heuristic possible-miss warnings for a result set.
detect_workflow: Detect which workflow recipe was followed and which steps were skipped.
dirty_files_for_change: Convenience: hoist git_dirty_files() for callers that want to share the same set between finalize_file_scores and a downstream stale_symbols scan. Just re-exports the underlying helper for API discoverability.
discover_symbol_ownership: Discover the likely owner of a symbol from git blame + doc-comment annotations.
drop_empty_recursive: Recursive variant of drop_empty_top_level. Walks every nested object and drops null/[]/{} children. Arrays’ elements are recursed into (an object inside an array still gets cleaned) but the array itself is preserved at its length — position often matters in arrays.
drop_empty_top_level: Token economy: drop top-level fields whose value is null, [], or {}. The agent doesn’t need the absence-of-X echo when X has no signal — they can infer “no entries” / “no warning” / “fresh” from absence.
drop_orphaned_edge_refs: Drop orphaned callee and caller refs whose target symbol_id no longer exists in the live index, in one commit. Returns the total number of individual refs dropped. Called by [hydrate_from_dir] after a hydrate pass to clean up any stale edges the sidecar may have carried.
edges_from_tree: Flatten the on-disk data-flow registry (from_symbol_id → [DataFlowEdge]) into a flat list. Malformed entries are skipped.
effect_detail_reason: Return a one-line explanation of an effect declaration’s verification state.
emit_audit: Fire-and-forget audit emit. Logs sink failures to stderr but never propagates — audit issues must never block the user’s operation.
endpoints_from_tree: Flatten the on-disk endpoint registry tree (contract_hash → repo_id → symbol_id → ServiceEndpoint) into a flat list. Malformed entries are skipped.
endpoints_report: Cross-service endpoints detected in this repo plus the in-repo matched edges. This is the JSON body asd endpoints --agent prints.
estimate_tokens: Estimate token count from a JSON string (rough: 1 token ≈ 4 chars).
explain_conflict_risk: Detect conflict risk for a single file. Returns Some(reason) if the file has unresolved merge markers or git status modifications; None when the file is clean.
explain_feedback_impacts: For each qname in qnames, determine which feedback verdict (if any) affected it and return a map of qname → FeedbackImpact.
explain_match: Explain why a symbol was returned for a query.
extract_summary
federated_edges: Federated edges over a set of per-repo manifests: match_edges on the union of every manifest’s endpoints. Each edge carries cross_repo (consumer and producer in different repos). With cross_repo_only, in-repo edges are dropped — the federated view a Team-tier tool wants: “a client in repo A calls a route served by repo B” (Plan Q t-005).
fetch_all_test_file_paths: Pre-fetch all test-tier file paths in a single DB open. Pass the result to test_files_for_source for per-file stem matching, avoiding repeated DB opens when processing many source files in a loop.
file_score_floor
finalize_file_scores: Sort + cliff-cut + build the likely_edit_files JSON array.
find_candidates: Return the top-depth (score, qname) pairs for a query.
find_covering_tests: Find test symbols in the FTS index that are likely to cover impl_qname.
find_indexed_test_files: Look up real indexed test files that correspond to source_file.
format_age: Format a unix timestamp age as a human-readable string (“3h ago”, “just now”).
gather_recency: Run one git log pass covering up to scan_commits commits and return a map of relative file path → FileRecency.
git_dirty_files: Return source file paths that are modified (dirty) since the last commit.
glob_match: Match a file path against a glob pattern.
http_contract: Build a normalized HTTP contract key from a method and a path (or full URL). A client call and a server route for the same endpoint must produce the same string: the method is upper-cased, any scheme+host is stripped, path parameters ({id}, :id, <id>, <int:id>) collapse to {}, query and fragment are dropped, and trailing slashes are removed.
hybrid_boost: Rust-side score boost applied on top of BM25 after FTS returns hits.
hydrate_from_dir: Read a .asd/v1/ sidecar under dir and write its contents back into the ASG repo via the existing stores.
hydrate_from_dir_with_cache: hydrate_from_dir, also writing the ledger entries it loads into the ledger cache, which answers list_entries.
in_memory_score: Score a symbol against query tokens using simple substring matching. Used when the FTS index is not populated.
in_memory_score_with_text: Same scoring rules as in_memory_score, but takes the symbol’s lowercased ledger-summary text directly so bulk callers can supply it from one ledger-tree walk instead of a per-symbol list_entries read (Plan T scale finding).
in_memory_scored_symbols: Bulk in-memory scorer shared by every fallback-search caller (core find_candidates, MCP code_search, CLI asd search): ONE by-qname tree read — deserializing the Symbol values the tree already contains — plus ONE ledger-tree walk, instead of a per-qname symbol fetch and a per-symbol ledger read (~54ms each on a cold FTS cache — ~9 minutes at 10k symbols while holding the engine mutex; Plan T scale finding).
intent_focus: One-line agent guidance for each intent.
intent_layer_order: Return the preferred layer display order for an intent. The standard order is used for unlisted layers.
is_doc_file: Returns true if this path should be offered to document adapters. Used to quickly gate file walking before reading content.
is_stopword: Returns true if the token is a stopword.
kind_str
ledger_counts: Count ledger entries in the cache, ASG and (when sidecar_root is given and has one) the sidecar, and the gaps between them.
list_all_effect_decls: List every stored EffectDecl in the workspace as (symbol_id, decl) pairs by walking the /asd/v1/effects subtree. Read-only. Used by overview-style consumers (e.g. asd-serve’s /api/v1/effects/overview) that need the whole distribution rather than one symbol’s decl — the per-symbol path stays EffectStore::get_effects.
load_exclude_sets: Plan J t-011: load named exclude sets from .asd/scopes.toml.
load_layer_overrides: Load user-defined layer overrides from .asd/layers.toml next to db_path.
load_scope_aliases: Load named scope aliases from .asd/scopes.toml in the current directory.
mark_fresh_reset: Write the meta/fresh-reset sentinel so agents know a deliberate reset occurred. Call this when asd init --reset or an equivalent wipe operation runs.
match_edges: Pair outbound endpoints with inbound endpoints sharing a contract key, producing cross-service edges. Endpoints may come from any repo (local + imported manifests). Edge confidence is the weaker of the two detections; cross_repo is set when the ends live in different repos.
matches_any_path_glob: Return true if file matches any of the given glob patterns.
missing_ledger_entries: Cached ledger entries whose entry_id appears nowhere in the ASG ledger tree. Matching by id rather than path means an entry moved by ledger rebind (whose cache row follows it to the new symbol) is never mistaken for a lost one.
normalize_repo_id: Normalize a git remote URL into a stable repo_id. SCP-form and HTTPS-form URLs for the same repo collapse to one id:
outdated_hooks: A warning when the project under root runs git hooks from before Plan B, which reload the whole ledger from .asd/v1/ with asd hydrate on every checkout and merge. On a store that holds more than the sidecar — anything since the last asd sync — that rewrote every entry each time, and before v1.4.7 it re-filed moved entries under their old symbols. asd init installs the current hooks, which import committed conclusions instead.
parse_intent: Parse and validate an intent string. Returns a static str or None if the value is unrecognised.
parse_params: Extract positional parameter names from a function signature, e.g. "def get_user(id, name=None, *args)" → ["id", "name"]. Skips *args/**kwargs and the self/cls receiver. Language-agnostic enough for the common name(params) shape.
parse_query: Parse a query string into (positive_tokens, exclusion_terms).
propagate_caller_invariants: Plan J t-001 caller-invariant propagation.
propagate_transitive: Compute transitive effects for each symbol in symbol_ids and write them back via effects.put_effects(...). Returns the count of symbols whose EffectDecl.transitive actually changed.
propose_test_path
propose_test_stub: Plan J t-007: language-aware test stub body for an agent that needs to write a new test against source_file exercising symbol_name. Returns the recommended skeleton — function declaration, arrange/act/assert comments, and a NotImplementedError-equivalent marker so the test FAILS until the agent fills it in (failing-first matches the missing-test recipe item which says “fails before the edit, passes after”).
prune_sidecar: Remove orphaned .asd/v1/ sidecar files — files whose keys no longer exist in the live ASG index. Returns the number of files/dirs removed.
pubsub_contract: Build a normalized pub-sub contract key from a topic / queue / event name.
query_tokens: Tokenise a free-form query into lowercase, stopword-filtered terms.
read_jsonl: Read back events from a JSONL file. Shared by OSS read-only consumers (asd audit tail) and the Pro verify_chain code path.
rebind_ledger: Move the ledger entries filed under each from symbol id onto its target symbol, in one commit.
repair_asg: Scan the ASG for integrity issues and, when dry_run is false, apply all auto-fixable corrections.
repair_asg_with_cache: repair_asg, also pointing the ledger cache at the copy kept of each duplicated entry — the cache answers list_entries, and a row left on a removed copy would hide the entry from its symbol.
resolve_edge: Resolve a detected data-flow site into a DataFlowEdge, given a way to look up a qname’s symbol_id and a callee’s parameter list. Returns None when the caller or callee can’t be resolved, or the arg position has no corresponding parameter.
resolve_exclude_set: Resolve a --exclude-set name to negative path globs. Unknown names return an empty list rather than treating the name as a glob (avoids silently turning a typo into a path filter).
resolve_repo_id: Resolve this repo’s id: an explicit override wins; otherwise the normalized origin git remote; otherwise the working-directory name as a last resort.
resolve_scope: Resolve a --scope name to path globs. Falls back to treating the scope name itself as a path glob if not in the map.
restore_missing_ledger_entries: Write every cache-only ledger entry back into ASG, with its reverse-index record, exactly as the cache holds it. Each restore is its own commit so the audit trail says what was recovered and from where.
result_bucket: Classify a result into a semantic bucket.
run_index: Run the full index pipeline over path.
scan_asg: Read-only integrity scan. Returns every detected RepairIssue without modifying any ASG state.
scan_sidecar: Scan .asd/conclusions/ for files and directories that don’t match the Plan K sidecar inclusion rule:
score_evidence_quality: Score task-closure evidence quality from available signals.
sidecar_lifecycle_state: Inspect the on-disk sidecar under dir and return its lifecycle state.
stale_warning: Return a stale-index warning string if the index is older than threshold_secs (default: 3600 = 1 hour), or None if fresh.
stale_warning_classified
suggest_better_queries: Suggest more specific queries when the current query is too broad.
suggest_scoped_queries: Suggest refined, scoped queries when the current query contains ambiguous terms.
symbol_fingerprint: Content-addressed fingerprint of a symbol’s body. Changes on every edit.
symbol_tier: Classify a source file into a symbol tier.
sync_to_dir: Mirror live ASG state into the .asd/v1/ sidecar under dir.
test_files_for_source: Filter all_test_files (from fetch_all_test_file_paths) for those whose stem matches source_file’s stem. Same logic as find_indexed_test_files but avoids the per-call DB open.
trim_for_agent: Trim a JSON value for LLM consumption.

Type Aliases§

FileScoreTuple: Per-row file score tuple shape used by the prepare-change pipeline. (score, file, layer, last_touched_days, hot, top_symbol, why)
Result
SymbolTier: Symbol tier — controls default inclusion and scoring penalty.

Crate agentstatedeveloper

Crate agentstatedeveloper Copy item path

Modules§

Structs§

Enums§

Constants§

Traits§

Functions§

Type Aliases§

Crate agentstatedeveloper