yoni-at-strix · Sep 25, 2026 · Sep 25, 2026 · Sep 25, 2026 · Sep 25, 2026 · Sep 25, 2026
diff --git a/strix/agents/prompts/system_prompt.jinja b/strix/agents/prompts/system_prompt.jinja
@@ -153,8 +153,9 @@ WHITE-BOX TESTING (code provided):
 - Local execution, unit/integration testing, patch verification, and HTTP requests against locally started in-scope services are normal authorized white-box validation
 - If dynamically running the code proves impossible after exhaustive attempts, pivot to comprehensive static analysis.
 - Try to infer how to run the code based on its structure and content.
-- Derive the code fix as PART OF reporting, not as a separate later pass: create_vulnerability_report already requires the concrete patch inline (`code_locations` with verbatim `fix_before`/`fix_after` and `fix_pr_body`), so the reporting agent that analyzes the root cause is the one that produces the fix. Do NOT spawn a downstream agent afterwards to re-derive/re-apply the same patch.
-- If you also apply and verify the patch in the repo (edit the file, re-test that the vulnerability is gone), do it in the same agent/turn while the analysis is fresh — right before or as part of filing the report — never as a second re-analysis pass.
+- Draft the initial fix candidate when you file the report. Use `code_locations` with verbatim `fix_before` and `fix_after`, plus `fix_pr_body`.
+- Treat the inline changes as a candidate, not as a completed fix. A later preparation stage can inspect and modify any required repository file.
+- Record checks you ran in `fix_verification`. Do not describe reasoned checks as executed checks.
 
 COMBINED MODE (code + deployed target present):
 - Treat this as static analysis plus dynamic testing simultaneously
@@ -240,7 +241,9 @@ VALIDATION REQUIREMENTS:
 - THREAT MODEL: before you start testing, call `get_threat_model` on the target you were pointed at — it is the scan's shared answer to who the attacker is, where the trust boundaries sit, and what counts as critical here. It is scoped to this scan and nothing carries over from an earlier run, so `found: false` means no agent on this run has derived one yet. Read it instead of re-deriving trust boundaries yourself; where your testing disproves it — a boundary it calls trusted turns out to be attacker-reachable, a role it did not know about, a host or endpoint it never listed — record that with `amend_threat_model` so the agents after you inherit the correction. Amending is not optional politeness: a model nobody corrects turns the first agent's guesses into everyone's assumptions.
 - Before filing any report, run the counterevidence pass: argue the strongest case AGAINST the finding, record what you found in the `counterevidence` field, set `confidence` honestly (a static-only trace you couldn't execute is at best `medium`), and state what evidence would change the severity. See the counterevidence and severity-calibration knowledge above.
 - A vulnerability is ONLY considered reported when a reporting agent uses create_vulnerability_report (or create_dependency_report for known-CVE dependency/supply-chain findings) with full details. Mentions in agent_finish, finish_scan, or generic messages are NOT sufficient
-- Reporting and fixing are ONE step, not two: when source is available, the reporting agent derives the concrete fix and files it INLINE via create_vulnerability_report (`code_locations` with `fix_before`/`fix_after` + `fix_pr_body`) — the report is not complete without it. Do NOT report first and then spawn a separate downstream agent to re-derive and re-apply the same patch; that just re-does the analysis and wastes tokens. (Do not silently patch a finding WITHOUT filing a report — the report, with its embedded fix, is the deliverable.)
+- When source is available, the reporting agent files an initial fix candidate with the report. The candidate uses `code_locations` with `fix_before` and `fix_after`, plus `fix_pr_body`.
+- Do not treat the candidate as a prepared fix. The fix preparation stage applies, repairs, tests, and independently verifies it after finding discovery closes.
+- Do not silently patch a finding without filing a report.
 - DEDUPLICATION: The create_vulnerability_report tool uses LLM-based deduplication. If it rejects your report as a duplicate, DO NOT attempt to re-submit the same vulnerability. Accept the rejection and move on to testing other areas. The vulnerability has already been reported by another agent. If your evidence proves more than the finding it matched (a working exploit where that one had only a static trace, a chain that raises the impact), revise that finding with update_vulnerability_report using the duplicate_of id — never re-file it.
 - HTTP EVIDENCE: a finding you validated through the proxy is not fully filed until `http_exchange_ids` carries the proxy request ids of the exchanges that prove it — the request that triggers the vulnerability plus the baseline/control request it differs from (an unauthenticated success next to the authenticated one, the payload response next to the benign one). Copy the ids exactly as `list_requests`/`view_request` show them, never invent or guess one, and never omit the field to bypass validation. Leave it out only when there is no captured HTTP exchange at all (static-only code findings, dependency CVEs). If you filed before the proving exchanges existed, attach them afterwards with update_vulnerability_report. Without the ids, the finding ships as prose nobody can replay.
 - REVISING A FINDING: use update_vulnerability_report (report id + the fields you want to replace + update_reason) when you learn something a finding already on file does not carry — you built the PoC after filing it, a chain raised its impact, further testing weakened it, or its counterevidence/remediation/code locations were wrong. Editing a finding needs no duplicate verdict, and it is always better than filing a second report for the same issue. Read the finding first with get_report, and pass only the fields that change.
@@ -360,7 +363,7 @@ ROOT AGENT ROLE:
 
 1. **CREATE AGENTS SELECTIVELY** - Spawn subagents when delegation materially improves parallelism, specialization, coverage, or independent validation. Deeper delegation is allowed when the child has a meaningfully different responsibility from the parent. Do not spawn subagents for trivial continuation of the same narrow task.
 2. **BLACK-BOX**: Discovery → Validation → Reporting (3 agents per vulnerability)
-3. **WHITE-BOX**: Discovery → Validation → Reporting-with-fix (3 agents per vulnerability — the reporting agent derives and files the fix inline; do NOT add a separate fixing agent that re-derives the same patch)
+3. **WHITE-BOX**: Discovery → Validation → Reporting with an initial fix candidate. The later preparation phase repairs and verifies the candidate.
 4. **MULTIPLE VULNS = MULTIPLE CHAINS** - Each vulnerability finding gets its own validation chain
 5. **CREATE AGENTS AS YOU GO** - Don't create all agents at start, create them when you discover new attack surfaces
 6. **ONE JOB PER AGENT** - Each agent has ONE specific task only
@@ -379,7 +382,7 @@ BLACK-BOX (domain/URL only):
 WHITE-BOX (source code provided):
 - Found authentication code issues? → Create authentication analysis agent
 - Auth agent finds potential vulnerability? → Create "Auth Validation Agent"
-- Validation agent confirms vulnerability? → Create "Auth Reporting Agent" that files the report AND its inline fix (`code_locations` + `fix_pr_body`) in one shot — no separate fixing agent
+- Validation agent confirms vulnerability? → Create "Auth Reporting Agent" that files the report and an initial fix candidate (`code_locations` + `fix_pr_body`)
 
 VULNERABILITY WORKFLOW (MANDATORY FOR EVERY FINDING):
 
@@ -401,10 +404,10 @@ Authentication Code Agent finds weak password validation
 Spawns "Auth Validation Agent" (proves it's exploitable)
     ↓
 If valid → Spawns "Auth Reporting Agent" (creates the vulnerability report
-           WITH the fix inline: code_locations fix_before/fix_after + fix_pr_body,
-           applying/verifying the patch in the same turn if desired)
+           with the initial fix candidate: code_locations fix_before/fix_after
+           + fix_pr_body)
     ↓
-STOP - no separate fixing agent; the fix was derived once, at report time
+STOP - the scan preparation phase handles repository-wide repair and verification
 ```
 
 CRITICAL RULES:
@@ -440,7 +443,7 @@ FOCUS PRINCIPLES:
 REALISTIC TESTING OUTCOMES:
 - **No Findings**: Agent completes testing but finds no vulnerabilities
 - **Validation Failed**: Initial finding was false positive, validation agent confirms it's not exploitable
-- **Valid Vulnerability**: Validation succeeds, spawns a reporting agent that files the report with the fix inline (white-box) — no separate fixing agent
+- **Valid Vulnerability**: Validation succeeds. The reporting agent files the report and an initial fix candidate for white-box findings.
 
 PERSISTENCE IS MANDATORY:
 - Real vulnerabilities take TIME - expect to need 2000+ steps minimum

diff --git a/strix/fix/__init__.py b/strix/fix/__init__.py
@@ -0,0 +1,52 @@
+"""Verified fix preparation contracts and runtime."""
+
+from strix.fix.contracts import (
+    CandidateLocation,
+    CheckResult,
+    CheckStatus,
+    CommandSpec,
+    FileManifestEntry,
+    FixCandidateV1,
+    FixEdit,
+    FixPreparationRequestV1,
+    FixPreparationResultV1,
+    PreparationState,
+    ReproductionSpec,
+    SourceIdentity,
+    SourceIdentityKind,
+    VerificationDecision,
+    VerifierResult,
+    candidate_from_legacy_report,
+)
+from strix.fix.prepare import (
+    PreparationCancelledError,
+    PreparationContext,
+    PreparationPolicy,
+    build_git_manifest,
+    prepare_fix,
+)
+
+
+__all__ = [
+    "CandidateLocation",
+    "CheckResult",
+    "CheckStatus",
+    "CommandSpec",
+    "FileManifestEntry",
+    "FixCandidateV1",
+    "FixEdit",
+    "FixPreparationRequestV1",
+    "FixPreparationResultV1",
+    "PreparationCancelledError",
+    "PreparationContext",
+    "PreparationPolicy",
+    "PreparationState",
+    "ReproductionSpec",
+    "SourceIdentity",
+    "SourceIdentityKind",
+    "VerificationDecision",
+    "VerifierResult",
+    "build_git_manifest",
+    "candidate_from_legacy_report",
+    "prepare_fix",
+]
diff --git a/strix/fix/contracts.py b/strix/fix/contracts.py
@@ -0,0 +1,278 @@
+"""Versioned contracts for verified fix preparation."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+from enum import StrEnum
+from pathlib import PurePosixPath
+from typing import TYPE_CHECKING, Literal, cast
+
+from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
+
+
+if TYPE_CHECKING:
+    from collections.abc import Mapping
+
+
+class ContractModel(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+
+class SourceIdentityKind(StrEnum):
+    COMMIT = "commit"
+    ARCHIVE = "archive"
+
+
+class PreparationState(StrEnum):
+    PREPARING = "preparing"
+    READY = "ready"
+    READY_WITH_GAPS = "ready_with_gaps"
+    NEEDS_REVIEW = "needs_review"
+    BLOCKED = "blocked"
+    FAILED = "failed"
+    STALE = "stale"
+
+
+class CheckStatus(StrEnum):
+    PASSED = "passed"
+    FAILED = "failed"
+    UNAVAILABLE = "unavailable"
+    CANCELLED = "cancelled"
+
+
+class VerificationDecision(StrEnum):
+    VERIFIED = "verified"
+    REJECTED = "rejected"
+    INCONCLUSIVE = "inconclusive"
+
+
+class SourceIdentity(ContractModel):
+    kind: SourceIdentityKind
+    value: str = Field(min_length=1)
+    repository: str | None = None
+
+    @field_validator("value")
+    @classmethod
+    def validate_value(cls, value: str) -> str:
+        normalized = value.strip().lower()
+        if not normalized or any(char not in "0123456789abcdef" for char in normalized):
+            raise ValueError("source identity must be a hexadecimal digest")
+        return normalized
+
+    @model_validator(mode="after")
+    def validate_digest_length(self) -> SourceIdentity:
+        valid_lengths = {40, 64} if self.kind is SourceIdentityKind.COMMIT else {64}
+        if len(self.value) not in valid_lengths:
+            expected = "40 or 64" if self.kind is SourceIdentityKind.COMMIT else "64"
+            raise ValueError(f"{self.kind} identity must contain {expected} hexadecimal characters")
+        return self
+
+
+def _validate_relative_path(value: str) -> str:
+    path = value.strip()
+    parsed = PurePosixPath(path)
+    if not path or parsed.is_absolute() or ".." in parsed.parts or path.startswith("./"):
+        raise ValueError("path must be relative to the repository root")
+    return path
+
+
+class CandidateLocation(ContractModel):
+    file: str
+    start_line: int = Field(ge=1)
+    end_line: int = Field(ge=1)
+    snippet: str | None = None
+    label: str | None = None
+
+    _relative_file = field_validator("file")(_validate_relative_path)
+
+    @model_validator(mode="after")
+    def validate_range(self) -> CandidateLocation:
+        if self.end_line < self.start_line:
+            raise ValueError("end_line must be greater than or equal to start_line")
+        return self
+
+
+class FixEdit(CandidateLocation):
+    before: str = Field(min_length=1)
+    after: str
+    original_sha256: str | None = None
+
+
+class CommandSpec(ContractModel):
+    name: str = Field(min_length=1, max_length=120)
+    argv: list[str] = Field(min_length=1)
+    required: bool = True
+    timeout_seconds: int = Field(default=300, ge=1, le=3600)
+    cwd: str = "."
+
+    _relative_cwd = field_validator("cwd")(_validate_relative_path)
+
+    @field_validator("argv")
+    @classmethod
+    def validate_argv(cls, value: list[str]) -> list[str]:
+        if any(not argument or "\x00" in argument for argument in value):
+            raise ValueError("command arguments must be non-empty and cannot contain NUL")
+        return value
+
+
+class ReproductionSpec(ContractModel):
+    instructions: str = Field(min_length=1)
+    command: CommandSpec | None = None
+
+
+class ReportedCheck(ContractModel):
+    name: str
+    result: str
+    executed: bool = False
+
+
+class FixCandidateV1(ContractModel):
+    version: Literal["1"] = "1"
+    source_identity: SourceIdentity | None = None
+    security_invariant: str = Field(min_length=1)
+    finding_locations: list[CandidateLocation] = []
+    draft_edits: list[FixEdit] = []
+    reproduction: ReproductionSpec | None = None
+    reported_checks: list[ReportedCheck] = []
+    known_gaps: list[str] = []
+
+    def digest(self) -> str:
+        payload = json.dumps(
+            self.model_dump(mode="json"),
+            sort_keys=True,
+            separators=(",", ":"),
+        ).encode()
+        return hashlib.sha256(payload).hexdigest()
+
+
+class FixPreparationRequestV1(ContractModel):
+    version: Literal["1"] = "1"
+    organization_id: str | None = None
+    scan_id: str
+    finding_id: str
+    repository_id: str | None = None
+    candidate: FixCandidateV1
+    checks: list[CommandSpec] = []
+    max_repair_attempts: int = Field(default=2, ge=1, le=5)
+    timeout_seconds: int = Field(default=1800, ge=30, le=14400)
+    network_allowed: bool = False
+    credentials_allowed: list[str] = []
+
+
+class CheckResult(ContractModel):
+    name: str
+    argv: list[str]
+    status: CheckStatus
+    exit_code: int | None = None
+    duration_seconds: float = Field(ge=0)
+    output: str = ""
+    required: bool = True
+
+
+class VerifierResult(ContractModel):
+    decision: VerificationDecision
+    summary: str
+    security_invariant_closed: bool = False
+    reproduction_executed: bool = False
+    reproduction_summary: str | None = None
+    sibling_paths_reviewed: list[str] = []
+    preserved_behaviors: list[str] = []
+    gaps: list[str] = []
+
+
+class FileManifestEntry(ContractModel):
+    path: str
+    operation: Literal["add", "modify", "delete"]
+    original_sha256: str | None = None
+    resulting_sha256: str | None = None
+    artifact_ref: str | None = None
+
+    _relative_path = field_validator("path")(_validate_relative_path)
+
+
+class FixPreparationResultV1(ContractModel):
+    version: Literal["1"] = "1"
+    state: PreparationState
+    stop_reason: str
+    source_identity: SourceIdentity | None
+    candidate_digest: str
+    final_file_manifest: list[FileManifestEntry] = []
+    artifact_ref: str | None = None
+    changed_files: list[str] = []
+    diff_summary: str = ""
+    checks: list[CheckResult] = []
+    security_reproduction: CheckResult | None = None
+    verifier: VerifierResult | None = None
+    gaps: list[str] = []
+    attempts: int = Field(default=0, ge=0)
+    elapsed_seconds: float = Field(default=0, ge=0)
+    cost_usd: float | None = Field(default=None, ge=0)
+
+
+def candidate_from_legacy_report(
+    report: Mapping[str, object],
+    *,
+    source_identity: SourceIdentity | None = None,
+) -> FixCandidateV1 | None:
+    raw_locations = report.get("code_locations")
+    if not isinstance(raw_locations, list):
+        return None
+    location_values = cast("list[object]", raw_locations)
+
+    locations: list[CandidateLocation] = []
+    edits: list[FixEdit] = []
+    for raw_value in location_values:
+        if not isinstance(raw_value, dict):
+            continue
+        raw = cast("dict[str, object]", raw_value)
+        file_path = raw.get("file")
+        start_line = raw.get("start_line")
+        end_line = raw.get("end_line")
+        if not isinstance(file_path, str) or type(start_line) is not int:
+            continue
+        if type(end_line) is not int:
+            end_line = start_line
+        common = {
+            "file": file_path,
+            "start_line": start_line,
+            "end_line": end_line,
+            "snippet": raw.get("snippet") if isinstance(raw.get("snippet"), str) else None,
+            "label": raw.get("label") if isinstance(raw.get("label"), str) else None,
+        }
+        try:
+            locations.append(CandidateLocation.model_validate(common))
+            before = raw.get("fix_before")
+            after = raw.get("fix_after")
+            if isinstance(before, str) and isinstance(after, str):
+                edits.append(FixEdit.model_validate({**common, "before": before, "after": after}))
+        except ValueError:
+            continue
+
+    if not locations:
+        return None
+
+    invariant = str(
+        report.get("remediation_steps") or report.get("technical_analysis") or ""
+    ).strip()
+    if not invariant:
+        invariant = "Resolve the reported security finding without changing legitimate behavior."
+
+    reported_checks: list[ReportedCheck] = []
+    verification = str(report.get("fix_verification") or "").strip()
+    if verification:
+        reported_checks.append(
+            ReportedCheck(name="reporting-agent verification", result=verification, executed=False)
+        )
+
+    return FixCandidateV1(
+        source_identity=source_identity,
+        security_invariant=invariant,
+        finding_locations=locations,
+        draft_edits=edits,
+        reproduction=ReproductionSpec(
+            instructions=str(report.get("poc_description") or report.get("evidence") or invariant)
+        ),
+        reported_checks=reported_checks,
+        known_gaps=["The reporting-agent verification is not independent."],
+    )