Evidence Graph Review
Scope and Authorization Gate
MUST confirm scope.md before any analysis step:
scope.md must exist under the current case root
auth.status must be granted
network_profile.mode must be set (offline | lab_only | authorized_target_only | unrestricted_lab)
If scope.md is missing or auth.status != granted, STOP and report scope deficiency. Do not proceed with any analysis or tool invocation.
Evidence Discipline
Evidence Records (E-xxx)
- Each evidence record is a self-contained Markdown file under the case root
- Must contain:
title, observed_at, source_type, source_ref, repro_command, raw_excerpt, linked_workitem, supersedes
content_hash and artifact_path are optional but MUST be recorded if a local artifact is referenced
repro_command must be a real command that a third party can run, or explicitly marked n/a with documentation of the offline limitation
Finding Records (F-xxx)
- Every Finding must reference at least one Evidence ID in
evidence_ids
- When
status=validated, confidence must be high or medium; low is only allowed with an explicit residual_risk note
- Findings without linked Evidence must be marked
status=candidate and promoted only after additional Evidence is gathered
Finding Sufficiency (R4*)
preliminary / candidate → >=1 Evidence (unchanged)
validated → SHOULD >=2 independent Evidence (best: 1 static + 1 dynamic). A single Evidence alone MUST NOT silently promote to validated — keep candidate, or record residual risk + human confirm
blocked promotion → record E-insufficient-evidence
Traceability
- All
F-xxx records must have evidence_ids pointing to actual E-xxx records in the same case root
P-xxx (Path) records must reference at least one E-xxx per step
- When reviewing scope, evidence, findings, or paths, the review is read-only and does not modify the case or any target
Confidence Grading
high → Evidence from multiple independent sources or a single well-anchored static/dynamic source
medium → One source with corroborating context, or secondary source confirming a candidate
low → single weak signal, cannot stand alone for validated promotion; residual_risk must be documented
Path Contract
Path Records (P-xxx)
path_type: attack | callflow | solve
- Each step must have
action and either evidence or finding reference
start and goal must be concrete (file:line, URL, or well-scoped description)
residual_risks must be listed if any step has uncertain outcomes
Evidence Chain in Generated Reports
When docs-generator produces a report, it MUST include:
- Scope summary linked to case
scope.md
- Evidence table or chapter
- Findings list (with
evidence_ids)
- At least one Path (attack/callflow/solve) with steps, evidence, and finding references
- Timeline summary (optional, linkable to
timeline.md)
Stop and Replan Conditions
Stop and report if any of the following hold:
auth.status in scope.md is not granted
- No
scope.md exists under the case root
- A Finding has no linked Evidence IDs after a scope sweep
- Confidence is
low without an explicit residual_risk note
- A Path step references an Evidence ID that does not exist in the same case root
- A hash verification failure is detected when
--verify-hashes --strict is requested
Replan if:
- Additional independent Evidence can be gathered without violating scope
- The user requests re-evaluation of confidence with new data
- A scope or activity boundary shift is requested (update
scope.md first)
Recommended Review Procedure
- Read
scope.md and confirm auth.status = granted
- List all
E-xxx records; confirm each F-xxx references at least one E-xxx
- Grade confidence per Finding
- Verify all
P-xxx steps have concrete start/goal and at least one evidence/finding reference
- If
--verify-hashes is requested, validate content_hash matches the artifact at artifact_path
- Generate a concise review summary: scope OK, evidence count, findings count, path count, hash status
- Stop: do not proceed to analysis, tool invocation, or report generation until all stop conditions are resolved
Minimal Viable Review Command (read-only)
python3 skills/case-review/scripts/review_case.py work/<case> --format markdown
Or, if Python is unavailable, a pure prose checklist with the same gate logic.
Version History
- 1.0: Initial release. Extracted from reverse-skill methodology docs (ops/scope-contract.md, ops/evidence-finding-path.md, ops/analysis-decision-framework.md). No scripts, no toolchain coupling.
1---2name: evidence-graph-review3description: Evidence Graph Review — audit evidence-to-finding-to-path traceability, confidence grading, scope gating, and stop/replan conditions when authorization or evidence quality is insufficient. Methodology-only; no scripts, no tool execution.4license: MIT5metadata:6 version: "1.0"7 source: https://skillmd.com/skills/jzkk720/evidence-graph-review8 upstream: reverse-skill (methodology extraction, no upstream toolchain coupling)9 port: >-10 Clean methodology-only port. No GPL/AGPL, no offensive toolchain, no client-specific script dependencies.11 Methodology extracted from reverse-skill: ops/scope-contract.md, ops/evidence-finding-path.md,12 ops/analysis-decision-framework.md. Rewritten as read-only checklist/playbook.13---1415# Evidence Graph Review1617## Scope and Authorization Gate1819**MUST** confirm `scope.md` before any analysis step:20- `scope.md` must exist under the current case root21- `auth.status` must be `granted`22- `network_profile.mode` must be set (offline | lab_only | authorized_target_only | unrestricted_lab)2324If `scope.md` is missing or `auth.status != granted`, **STOP** and report scope deficiency. Do not proceed with any analysis or tool invocation.2526## Evidence Discipline2728### Evidence Records (E-xxx)2930- Each evidence record is a self-contained Markdown file under the case root31- Must contain: `title`, `observed_at`, `source_type`, `source_ref`, `repro_command`, `raw_excerpt`, `linked_workitem`, `supersedes`32- `content_hash` and `artifact_path` are optional but MUST be recorded if a local artifact is referenced33- `repro_command` must be a real command that a third party can run, or explicitly marked `n/a` with documentation of the offline limitation3435### Finding Records (F-xxx)3637- Every Finding must reference at least one Evidence ID in `evidence_ids`38- When `status=validated`, `confidence` must be `high` or `medium`; `low` is only allowed with an explicit `residual_risk` note39- Findings without linked Evidence must be marked `status=candidate` and promoted only after additional Evidence is gathered4041### Finding Sufficiency (R4*)4243- `preliminary / candidate` → >=1 Evidence (unchanged)44- `validated` → SHOULD >=2 independent Evidence (best: 1 static + 1 dynamic). A single Evidence alone MUST NOT silently promote to validated — keep `candidate`, or record residual risk + human confirm45- `blocked promotion` → record `E-insufficient-evidence`4647### Traceability4849- All `F-xxx` records must have `evidence_ids` pointing to actual `E-xxx` records in the same case root50- `P-xxx` (Path) records must reference at least one `E-xxx` per step51- When reviewing scope, evidence, findings, or paths, the review is read-only and does not modify the case or any target5253### Confidence Grading5455- `high` → Evidence from multiple independent sources or a single well-anchored static/dynamic source56- `medium` → One source with corroborating context, or secondary source confirming a candidate57- `low` → single weak signal, cannot stand alone for validated promotion; residual_risk must be documented5859## Path Contract6061### Path Records (P-xxx)6263- `path_type`: `attack` | `callflow` | `solve`64- Each step must have `action` and either `evidence` or `finding` reference65- `start` and `goal` must be concrete (file:line, URL, or well-scoped description)66- `residual_risks` must be listed if any step has uncertain outcomes6768### Evidence Chain in Generated Reports6970When `docs-generator` produces a report, it MUST include:711. Scope summary linked to case `scope.md`722. Evidence table or chapter733. Findings list (with `evidence_ids`)744. At least one Path (attack/callflow/solve) with steps, evidence, and finding references755. Timeline summary (optional, linkable to `timeline.md`)7677## Stop and Replan Conditions7879**Stop** and report if any of the following hold:80811. `auth.status` in `scope.md` is not `granted`822. No `scope.md` exists under the case root833. A Finding has no linked Evidence IDs after a scope sweep844. Confidence is `low` without an explicit `residual_risk` note855. A Path step references an Evidence ID that does not exist in the same case root866. A hash verification failure is detected when `--verify-hashes --strict` is requested8788**Replan** if:89901. Additional independent Evidence can be gathered without violating scope912. The user requests re-evaluation of confidence with new data923. A scope or activity boundary shift is requested (update `scope.md` first)9394## Recommended Review Procedure95961. Read `scope.md` and confirm `auth.status = granted`972. List all `E-xxx` records; confirm each `F-xxx` references at least one `E-xxx`983. Grade confidence per Finding994. Verify all `P-xxx` steps have concrete `start`/`goal` and at least one evidence/finding reference1005. If `--verify-hashes` is requested, validate `content_hash` matches the artifact at `artifact_path`1016. Generate a concise review summary: scope OK, evidence count, findings count, path count, hash status1027. Stop: do not proceed to analysis, tool invocation, or report generation until all stop conditions are resolved103104## Minimal Viable Review Command (read-only)105106```bash107python3 skills/case-review/scripts/review_case.py work/<case> --format markdown108```109110Or, if Python is unavailable, a pure prose checklist with the same gate logic.111112---113## Version History114115- 1.0: Initial release. Extracted from reverse-skill methodology docs (ops/scope-contract.md, ops/evidence-finding-path.md, ops/analysis-decision-framework.md). No scripts, no toolchain coupling.