fix(events,tools,toolintent): failure attribution, one path normalization rule, file_copy (#713)

Three generic harness fixes from the web-ui postmortem dataset. Nothing here keys
on a language, framework, build tool or task type.

1. Failure attribution. WorkflowFailedEvent carries one primary FailureAttribution
   (AGENT | HARNESS | WORKFLOW | ENVIRONMENT | PROVIDER | OPERATOR | UNKNOWN),
   defaulted to UNKNOWN so pre-field events replay unchanged. FailureAttributor is
   the deterministic reason->layer mapping, used both at emission and when
   classifying history, so the baseline and the live metric are one measurement.
   Emission sites set it: failWorkflow derives from the reason unless the caller
   knows the layer, cancellation is OPERATOR, the server catch-all falls back to
   HARNESS, a grounding-rejected plan is AGENT. Multi-cause chains stay on
   FailureTicketOpened — no second causal structure.

   GET /metrics/failure-attribution (FailureAttributionInspectionService, mirroring
   ToolReliabilityInspectionService) reports counts, share, UNKNOWN share, the
   preserved reasons and the ticket categories from the same sessions. Read-only:
   historical events are classified at READ time and reported as `inferred`, never
   written back over an append-only log.

   Baseline over the local log, 122 terminal failures: AGENT 51 (41.8%),
   OPERATOR 28 (23.0%), WORKFLOW 19 (15.6%), PROVIDER 15 (12.3%), HARNESS 6 (4.9%),
   ENVIRONMENT 3 (2.5%), UNKNOWN 0.

2. The `~` guard bug. ToolPath is now the ONE canonical normalization rule
   (expand a leading `~`/`~/`, keep absolutes, anchor relatives on the session
   working dir). Every filesystem tool, all six plane-2 path rules and the approval
   preview resolve through it, so policy and existence checks inspect the path the
   tool will operate on. `~/.gradle/init.d/offline.gradle` used to resolve to
   `<workspace>/~/.gradle/...`: reported non-existent AND in-workspace, so the
   reference gate called a real file a hallucination and the out-of-workspace prompt
   never fired. Containment and external-read approval behaviour are unchanged —
   the expanded path is simply outside the workspace, where it always belonged.

3. file_copy (#713). A first-class tool with the writer's jail, tier, receipt,
   replay and CAS pre/post images; static and binary assets no longer move through
   the model's token stream. Needed one generic split: ParamRole.SOURCE_PATH marks a
   path a call reads FROM, so containment gates judge both params while write-target
   gates (read-before-write, stale-write, write scope, write manifest) judge the
   mutated one. ReadBeforeWriteRule exempts any call declaring a SOURCE_PATH: its
   content comes from disk, not from memory, and requiring a read of a binary is
   unsatisfiable. Existing tools declare no SOURCE_PATH, so their behaviour is
   byte-identical.

Tests: ToolPathTest (9), FailureAttributionTest (10), PathNormalizationRuleTest (6),
FileCopyToolTest (10), plus a home-relative FileReadTool read. ./gradlew check green.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
2026-08-27 11:57:49 +04:00
parent 519290368f
commit 6a8a7b31c1
36 changed files with 1232 additions and 79 deletions
+3 -1
View File
@@ -24,7 +24,7 @@ CORREX kernel team. This module enforces Hard Invariant #9 for the tool-call pat
- `WorkspacePolicy` — aggregates rules and configuration for a workspace.
- `WorldProbe` — performs environment checks (filesystem, network) and records the observations as events immediately (Hard Invariant #9). Never call `WorldProbe` during replay.
- `EgressAllowlist` — current egress allowlist; rebuilt from `EgressAllowlistProjection` (in `core:events`).
- `ParamValueExtractor` — extracts typed parameter values from tool call arguments.
- `ParamValueExtractor` — extracts typed parameter values from tool call arguments. `candidatePathStrings` = every path-like argument (`ParamRole.PATH` + `ParamRole.SOURCE_PATH`), used by the containment/existence gates; `writeTargetPathStrings` = only the paths a call MUTATES (`ParamRole.PATH`), used by the write-target gates. A tool declaring none of those roles falls back to sniffing path-like strings, so `shell` is unaffected.
- `RiskMapping` — maps rule violations to risk levels for `core:risk`.
- `SessionContext` — session-scoped context passed to rules during evaluation.
@@ -33,6 +33,8 @@ CORREX kernel team. This module enforces Hard Invariant #9 for the tool-call pat
- Hard Invariant #9: all `WorldProbe` calls record observations as events. Replay reads those recorded events — it must not call `WorldProbe` again.
- Hard Invariant #5: every tool call must be assessed before execution. Assessment result is recorded as `ToolCallAssessmentEvents` in `core:events`.
- New rules implement `ToolCallRule` and are registered in `WorkspacePolicy`. Do not add rule logic directly to `ToolCallAssessor`.
- Resolve every model-supplied path through `ToolPath.resolve` (`core:tools`) — the one canonical normalization rule, shared with the filesystem tools. A rule that resolves paths itself will judge a different path than the tool operates on (the `~` bug: `~/x` resolved to `<workspace>/~/x`, so a real home-directory file was reported as a non-existent in-workspace file and the out-of-workspace prompt never fired).
- A call that declares a `SOURCE_PATH` takes its content from disk, not from the model, so `ReadBeforeWriteRule` exempts it: requiring a read of a copied file's bytes is unsatisfiable for binaries and defeats the purpose of the tool.
## Verification