docs(wiki): expand hermes attestation claim narratives and archive draft

This commit is contained in:
David Abutbul
2026-04-15 20:59:03 +00:00
parent 65b83d7da6
commit 1f1dde41bf
5 changed files with 300 additions and 132 deletions
@@ -1,111 +0,0 @@
# Hermes Attestation Guardian (Historical Draft Record)
Date: 2026-04-15
Owner: draft prepared for delegated implementation
Status: Implemented in repository (v0.0.1); this draft is historical context only
## 1) Proposed skill identity
Name: hermes-attestation-guardian
Why this name:
- Explicitly Hermes-scoped (no ambiguity with OpenClaw runtime skills)
- Keeps ClawSec "guardian" naming family
- Clear security function (attestation + guard)
Tagline:
"Continuously attest Hermes runtime security posture and surface drift before it becomes incident."
## 2) Product positioning (Hermes-first)
Primary audience:
- Operators running Hermes CLI/Gateway infra (Telegram/Matrix/Discord, local + remote hosts)
Non-goal:
- OpenClaw hook runtime protection
- Generic multi-agent platform abstraction beyond Hermes
Core value pillars:
1. Trust posture visibility: one signed snapshot of security state
2. Drift detection: compare baseline vs current and emit high-signal deltas
3. Operator-safe controls: no destructive remediation by default, explicit approvals required
## 3) Security outcomes
The skill should attest and diff at least:
- Enabled toolsets and risky toggles
- Messaging/gateway active state
- Signed feed verification state (where configured)
- Integrity-sensitive file hashes (operator-selected policy)
- Runtime version + dependency fingerprint summary
Priority alerts:
- unsigned mode or bypass flags enabled in production profile
- critical verification regression (signature/checksum failure)
- protected config/runtime file drift
- change in trust anchors (key fingerprint mismatch)
## 4) Hermes alignment rules (must-have)
Naming and metadata:
- skill.json name must be hermes-attestation-guardian
- platform field must be hermes
- triggers must include explicit Hermes language (avoid generic "security check" only)
Paths and side effects:
- default output under ~/.hermes/security/attestations/
- no writes to ~/.openclaw paths
- no destructive actions (quarantine/delete/restore) in MVP
Runtime/UX:
- fail closed for attestation verification errors
- operator-facing messages must separate: INFO / WARNING / CRITICAL
- remediation must be recommendation-only unless user explicitly opts in
## 5) ClawSec pipeline/distribution requirements
Based on existing repo automation, this skill must satisfy:
- skills/<name>/skill.json contains required fields: name, version, description, author, license, sbom.files
- skills/<name>/SKILL.md includes YAML frontmatter with matching version
- version parity: skill.json version == SKILL.md frontmatter version
- CHANGELOG.md must contain heading exactly:
## [<version>] - YYYY-MM-DD
- SBOM required file entries must exist in repo
- README.md recommended for site rendering
Release mechanics this skill should be compatible with:
- scripts/release-skill.sh version bump + URL rewrites
- .github/workflows/skill-release.yml parity/release-note checks
- utils/validate_skill.py local validation
## 6) Skeleton contents created
skills/hermes-attestation-guardian/
- skill.json
- SKILL.md
- README.md
- CHANGELOG.md
- scripts/generate_attestation.mjs
- scripts/verify_attestation.mjs
- scripts/setup_attestation_cron.mjs
- lib/diff.mjs
- test/attestation_schema.test.mjs
- test/attestation_diff.test.mjs
## 7) Delegate implementation brief
Delegate goal:
Prepare a production-ready v0.0.1 of hermes-attestation-guardian from this skeleton while keeping Hermes-only positioning and ClawSec release compatibility.
Required implementation scope:
1. Implement deterministic attestation schema + generator CLI
2. Implement verifier CLI (schema + signature/checksum hooks, fail-closed semantics)
3. Implement baseline diff engine with stable severity mapping
4. Implement cron setup helper for Hermes environments (no OpenClaw assumptions)
5. Add/expand tests for schema, diff correctness, and failure behavior
6. Keep docs and metadata aligned with actual behavior (no aspirational claims)
Acceptance checks before return:
- python utils/validate_skill.py skills/hermes-attestation-guardian
- Version parity and changelog heading satisfy skill-release policy
- Node tests in skill directory pass
- Tool/runtime side effects are explicitly documented in SKILL.md
+1 -1
View File
@@ -15,7 +15,7 @@
- Updated index and cross-links to use `wiki/` as the documentation source of truth.
- Added a dedicated module page for `clawsec-scanner` and linked it from `wiki/INDEX.md`.
- Future updates should preserve existing headings and append `Update Notes` sections when making deltas.
- 2026-04-15: Added `wiki/modules/hermes-attestation-guardian.md` documenting PR claims in human-friendly claim→wiring format.
- 2026-04-15: Expanded `wiki/modules/hermes-attestation-guardian.md` into full narrative claim breakdowns (people-speak + wiring + verification + scenario) and moved draft-plan context into `wiki/modules/hermes-attestation-guardian-draft-history.md`.
## Source References
- README.md
+3 -1
View File
@@ -31,6 +31,7 @@
- [ClawSec Suite Core](modules/clawsec-suite.md)
- [ClawSec Scanner](modules/clawsec-scanner.md)
- [Hermes Attestation Guardian](modules/hermes-attestation-guardian.md)
- [Hermes Attestation Guardian Draft History (Archived)](modules/hermes-attestation-guardian-draft-history.md)
- [NanoClaw Integration](modules/nanoclaw-integration.md)
- [Automation and Release Pipelines](modules/automation-release.md)
- [Local Validation and Packaging Tools](modules/local-tooling.md)
@@ -42,7 +43,7 @@
- [Generation Metadata](GENERATION.md)
## Update Notes
- 2026-04-15: Added Hermes Attestation Guardian module page with human-friendly claim-to-code/test mappings from PR description.
- 2026-04-15: Expanded Hermes Attestation Guardian module page into full narrative, claim-by-claim operator guidance (no claim tables), and added archived draft-history module page.
- 2026-03-10: Added ClawSec Scanner module documentation and linked it under Modules.
- 2026-02-26: Added Operations pages and updated navigation guidance after migrating root docs into wiki pages.
@@ -58,4 +59,5 @@
- skills/hermes-attestation-guardian/skill.json
- wiki/modules/clawsec-scanner.md
- wiki/modules/hermes-attestation-guardian.md
- wiki/modules/hermes-attestation-guardian-draft-history.md
- .github/workflows/ci.yml
@@ -0,0 +1,54 @@
# Module History: Hermes Attestation Guardian Draft (Archived)
## Purpose
This page preserves the original planning draft that led to `hermes-attestation-guardian` v0.0.1.
It is historical context, not current behavior contract.
## Status
- Draft date: 2026-04-15
- Current status: implemented in repository as `skills/hermes-attestation-guardian` v0.0.1
- Source of truth for live behavior: skill code, tests, and `wiki/modules/hermes-attestation-guardian.md`
## What the draft got right
- Hermes-only positioning (not OpenClaw hook runtime scope).
- Fail-closed verification as a core requirement.
- Deterministic attestation and digest binding requirements.
- Baseline-vs-current drift detection with severity ranking.
- Safe cron automation expectations (explicit apply, non-destructive defaults).
## Original design intent (summarized)
1) Identity and scope
- Name should clearly indicate Hermes scope and guardian role.
- Metadata should make platform targeting explicit.
2) Security outcomes
- Snapshot posture and integrity-sensitive inputs.
- Detect risky toggles, verification regressions, and trust/file drift.
- Prioritize high-signal alerts for operators.
3) Alignment rules
- Keep side effects under Hermes paths.
- Avoid destructive remediation in MVP.
- Keep operator-facing criticality clear.
4) Packaging/release compatibility
- Match ClawSec skill metadata and changelog requirements.
- Ensure local validation and test gates pass before release.
5) Delegate implementation scope
- Build generator, verifier, diff logic, cron helper, and tests.
- Keep docs aligned to implemented behavior.
## What changed from draft to implementation
- Implementation hardened path-scope checks (including symlink-aware escape defense).
- Verifier baseline trust was made explicit and fail-closed before diffing.
- Cron managed-marker parser hardened to fail closed on malformed marker structure.
- Wiki documentation now maps each PR claim to wiring and tests with human-readable operator guidance.
## Where to look now
- Live module documentation:
- `wiki/modules/hermes-attestation-guardian.md`
- Live skill implementation:
- `skills/hermes-attestation-guardian/`
- Validation tests:
- `skills/hermes-attestation-guardian/test/`
+242 -19
View File
@@ -2,38 +2,262 @@
## Responsibilities
- Produce a deterministic Hermes runtime security snapshot (attestation).
- Verify attestation integrity in fail-closed mode before trust decisions.
- Verify attestation integrity in fail-closed mode before any trust decision.
- Compare trusted baseline vs current posture and classify drift severity.
- Provide a safe, Hermes-scoped automation path for periodic attestation checks.
## Claims from PR description (human-friendly)
## PR Claims: Full Human-Friendly Breakdown
| Claim | In people-speak | Where it is wired (code/config) | How we verify it |
| --- | --- | --- | --- |
| Adds deterministic attestation generation with canonicalized payload digesting. | If nothing changed in Hermes, the attestation fingerprint should stay the same. No noisy diffs from JSON ordering. | `skills/hermes-attestation-guardian/scripts/generate_attestation.mjs`, `skills/hermes-attestation-guardian/lib/attestation.mjs` (`stableStringify`, canonical digest helpers), `skills/hermes-attestation-guardian/skill.json` | `node skills/hermes-attestation-guardian/test/attestation_schema.test.mjs` checks deterministic output and digest/schema expectations. |
| Enforces fail-closed verification for schema, digest, optional expected checksum, and detached signatures. | Verification must stop with an error when integrity checks fail; never “best effort pass.” | `skills/hermes-attestation-guardian/scripts/verify_attestation.mjs`, `skills/hermes-attestation-guardian/lib/attestation.mjs`, operator notes in `skills/hermes-attestation-guardian/skill.json` and `SKILL.md` | `node skills/hermes-attestation-guardian/test/attestation_cli.test.mjs` covers tamper/error paths and non-zero exits. |
| Adds baseline authenticity and drift-severity classification for risky toggles, feed verification regressions, trust anchor drift, and watched file drift. | Baseline diff is trusted only when baseline authenticity is proven, then drift is ranked by severity so operators can act fast. | Baseline trust gates in `scripts/verify_attestation.mjs`; drift engine in `lib/diff.mjs`; docs in `README.md`/`SKILL.md` | `node skills/hermes-attestation-guardian/test/attestation_cli.test.mjs` (baseline trust/tamper) and `node skills/hermes-attestation-guardian/test/attestation_diff.test.mjs` (severity mapping). |
| Adds Hermes-only cron setup helper with managed marker block and print-only default. | Cron setup is opt-in and safe by default: it prints planned changes unless `--apply` is explicit. | `skills/hermes-attestation-guardian/scripts/setup_attestation_cron.mjs`, `skills/hermes-attestation-guardian/SKILL.md` | `node skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs` validates print-only behavior and managed-block logic. |
| Includes output-scope/path guardrails for attestation artifacts and policy parsing safeguards. | Output writes are constrained to Hermes attestation scope, including symlink-aware escape defense. | `skills/hermes-attestation-guardian/lib/attestation.mjs` (`resolveHermesScopedOutputPath`), `scripts/generate_attestation.mjs`, `scripts/setup_attestation_cron.mjs` | `node skills/hermes-attestation-guardian/test/attestation_cli.test.mjs` includes out-of-scope and symlink-escape rejection checks. |
| Cron managed-block parser fails closed on malformed markers. | If cron markers are broken, updater refuses to rewrite instead of risking accidental deletion. | `skills/hermes-attestation-guardian/scripts/setup_attestation_cron.mjs` (`removeManagedBlock`) | `node skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs` covers dangling start, unmatched end, and nested marker failures. |
This section rewrites each PR claim as an operator-facing explanation, then ties it to exact code and tests.
### Claim 1: Adds deterministic attestation generation with canonicalized payload digesting.
Absolutely — in people-speak:
We create a security snapshot of Hermes in a way that is reproducible, then fingerprint it in a stable way so tampering or real drift is obvious.
What this means in practice:
1) Attestation generation
- Think of it as a report card for Hermes security posture at a moment in time.
- It records posture fields, trust anchors, watched-file hashes, and metadata.
2) Deterministic output
- Same state should produce the same attestation content.
- No noise from object insertion order or formatting randomness.
3) Canonicalization before hashing
- Payload is normalized into one canonical JSON representation.
- This removes ambiguity from normal JSON variations.
4) Digest binding
- SHA-256 is computed over canonical payload content.
- Any meaningful change to payload changes digest.
- Any post-generation tampering causes verification mismatch.
Where it is wired:
- `skills/hermes-attestation-guardian/scripts/generate_attestation.mjs`
- `skills/hermes-attestation-guardian/lib/attestation.mjs`
- `stableSortObject`
- `stableStringify`
- `sha256Hex`
- `buildAttestation`
- `computeCanonicalDigest`
- `validateDigestBinding`
How to verify:
- `node skills/hermes-attestation-guardian/test/attestation_schema.test.mjs`
- proves same-input determinism and canonical digest consistency.
- `node skills/hermes-attestation-guardian/test/attestation_cli.test.mjs`
- proves post-generation tamper causes fail-closed digest mismatch.
Quick scenario:
- Same state: run generator twice with unchanged inputs -> same digest.
- Tampered file: flip a posture value in JSON -> verifier fails on canonical digest mismatch.
---
### Claim 2: Enforces fail-closed verification for schema, digest, optional expected checksum, and detached signatures.
In people-speak:
Verification is not “best effort.” If a trust check fails, verification fails. No soft pass.
What is fail-closed here:
1) Schema must be valid.
2) Canonical digest must match payload.
3) If `--expected-sha256` is supplied, file bytes must exactly match.
4) If detached signature verification is requested, signature + public key must both be present and valid.
Where it is wired:
- `skills/hermes-attestation-guardian/scripts/verify_attestation.mjs`
- schema checks
- digest checks
- expected checksum check
- detached signature verification
- non-zero exit on critical failure
- `skills/hermes-attestation-guardian/lib/attestation.mjs`
- `validateAttestationSchema`
- `validateDigestBinding`
How to verify:
- `node skills/hermes-attestation-guardian/test/attestation_schema.test.mjs`
- proves schema rejection and digest algorithm validation behavior.
- `node skills/hermes-attestation-guardian/test/attestation_cli.test.mjs`
- proves tamper path exits non-zero (fail closed).
Quick scenario:
- CI pins expected SHA and requires detached signature.
- Artifact is modified or signed incorrectly -> verification exits non-zero and blocks pipeline.
---
### Claim 3: Adds baseline authenticity and drift-severity classification for risky toggles, feed verification regressions, trust anchor drift, and watched file drift.
In people-speak:
You only compare against a baseline after proving the baseline itself is authentic. Then differences are ranked by severity so operators can respond quickly.
What this gives operators:
1) Authenticated baseline gate
- Baseline must be trusted (pinned digest and/or detached signature trust path).
- Untrusted baseline is rejected before diffing.
2) Severity-ranked drift findings
- Critical/high/medium/low/info mapping instead of flat alerts.
- High-signal categories include:
- risky toggle enablement,
- feed verification regressions,
- trust anchor hash drift,
- watched file hash drift.
3) Policy-driven failure threshold
- Verification can fail when findings meet/exceed configured severity threshold.
Where it is wired:
- Baseline trust and diff orchestration:
- `skills/hermes-attestation-guardian/scripts/verify_attestation.mjs`
- Drift engine and severity mapping:
- `skills/hermes-attestation-guardian/lib/diff.mjs`
- `diffAttestations`
- `highestSeverity`
- `severityAtOrAbove`
How to verify:
- `node skills/hermes-attestation-guardian/test/attestation_cli.test.mjs`
- proves untrusted baseline rejection and digest-pinned baseline handling.
- `node skills/hermes-attestation-guardian/test/attestation_diff.test.mjs`
- proves classification for key drift types and highest-severity behavior.
Quick scenario:
- Yesterdays baseline is pinned and trusted.
- Today `allow_unsigned_mode` flips on and trust anchor hash changes.
- Diff emits critical findings and verifier can fail run by severity policy.
---
### Claim 4: Adds Hermes-only cron setup helper with managed marker block and print-only default.
In people-speak:
You get a scheduler helper that is safe by default: it shows planned cron changes first, and only writes when you explicitly ask.
What “safe by default” means:
1) Hermes-only framing in UX and docs.
2) Managed marker block for clean replacement of only this modules cron section.
3) Print-only default; write path requires explicit `--apply`.
Where it is wired:
- `skills/hermes-attestation-guardian/scripts/setup_attestation_cron.mjs`
- managed markers
- print-only defaults
- apply path
- Supporting scope/docs:
- `skills/hermes-attestation-guardian/SKILL.md`
- `skills/hermes-attestation-guardian/skill.json`
How to verify:
- `node skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs`
- proves Hermes-only messaging and managed-block behavior.
- proves default mode is preview-oriented and apply path is explicit.
Quick scenario:
- Operator runs cron helper without flags -> sees proposed block only.
- Operator reviews, then reruns with `--apply` -> only managed block is updated.
---
### Claim 5: Includes output-scope/path guardrails for attestation artifacts and policy parsing safeguards.
In people-speak:
Artifact writes are fenced into Hermes attestation scope, including symlink-escape defenses, and policy parsing is normalized/defensive so bad input fails cleanly.
What this protects against:
1) Out-of-scope writes
- Output path must remain under `HERMES_HOME/security/attestations`.
2) Symlink escapes
- Path resolution checks nearest existing ancestors and symlink behavior to prevent “write outside root” tricks.
3) Safer policy parsing
- Missing/invalid structure gets normalized defaults where appropriate.
- Malformed JSON fails closed.
- List fields are trimmed, deduplicated, and sorted.
Where it is wired:
- Guardrails:
- `skills/hermes-attestation-guardian/lib/attestation.mjs`
- `resolveHermesScopedOutputPath`
- Call sites:
- `skills/hermes-attestation-guardian/scripts/generate_attestation.mjs`
- `skills/hermes-attestation-guardian/scripts/setup_attestation_cron.mjs`
- Policy parsing:
- `skills/hermes-attestation-guardian/lib/attestation.mjs`
- `parseAttestationPolicy`
How to verify:
- `node skills/hermes-attestation-guardian/test/attestation_cli.test.mjs`
- proves out-of-scope and symlink-escape output rejection.
- `node skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs`
- proves cron helper also rejects out-of-scope output target.
Quick scenario:
- Operator accidentally sets `--output /tmp/current.json`.
- Tool exits with critical path-scope error instead of writing outside Hermes scope.
---
### Claim 6: Cron managed-block parser fails closed on malformed markers.
In people-speak:
If cron markers are malformed (dangling start/end or nested blocks), updater refuses to rewrite crontab to avoid accidental deletion or corruption.
What this means operationally:
1) Marker structure is treated as integrity-sensitive input.
2) Malformed structure throws and aborts apply path.
3) No crontab write occurs after malformed marker detection.
Where it is wired:
- `skills/hermes-attestation-guardian/scripts/setup_attestation_cron.mjs`
- `removeManagedBlock`
- marker parsing and malformed-marker throw paths
How to verify:
- `node skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs`
- proves fail-closed behavior for:
- dangling start marker,
- unmatched end marker,
- nested markers,
- and verifies no write on malformed input.
Quick scenario:
- Existing crontab has managed start marker with no end marker.
- Running `--apply` aborts with malformed-marker error and leaves crontab unchanged.
## Key Files
- `skills/hermes-attestation-guardian/skill.json`: metadata, platform scope, operator review notes, SBOM.
- `skills/hermes-attestation-guardian/SKILL.md`: operator playbook, CLI usage, fail-closed policy.
- `skills/hermes-attestation-guardian/README.md`: quickstart and practical behavior notes.
- `skills/hermes-attestation-guardian/lib/attestation.mjs`: canonicalization, digest binding, schema checks, scoped output resolution.
- `skills/hermes-attestation-guardian/lib/attestation.mjs`: canonicalization, digest binding, schema checks, scoped output resolution, policy parsing.
- `skills/hermes-attestation-guardian/lib/diff.mjs`: baseline drift comparison and severity classification.
- `skills/hermes-attestation-guardian/scripts/generate_attestation.mjs`: deterministic attestation generation CLI.
- `skills/hermes-attestation-guardian/scripts/verify_attestation.mjs`: fail-closed verifier and baseline trust enforcement.
- `skills/hermes-attestation-guardian/scripts/setup_attestation_cron.mjs`: cron managed-block helper.
## Public Interfaces
| Interface | Consumer | Behavior |
| --- | --- | --- |
| `generate_attestation.mjs` CLI | Operators/automation | Creates canonicalized attestation JSON and optional checksum artifact. |
| `verify_attestation.mjs` CLI | Operators/automation/cron | Enforces schema/digest/signature checks and optional trusted baseline drift checks. |
| `setup_attestation_cron.mjs` CLI | Operators | Prints or applies managed cron block for scheduled generate+verify runs. |
| Diff output contract | Operators/CI | Emits severity-ranked drift findings for security triage. |
- `generate_attestation.mjs` CLI
- Consumer: operators/automation
- Behavior: creates canonicalized attestation JSON and optional checksum artifact.
- `verify_attestation.mjs` CLI
- Consumer: operators/automation/cron
- Behavior: enforces schema/digest/signature checks and optional trusted-baseline drift checks.
- `setup_attestation_cron.mjs` CLI
- Consumer: operators
- Behavior: prints or applies managed cron block for scheduled generate+verify runs.
- Diff output contract
- Consumer: operators/CI
- Behavior: emits severity-ranked drift findings for security triage.
## Validation Commands
```bash
@@ -45,7 +269,7 @@ node skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs
```
## Update Notes
- 2026-04-15: Added module page for `hermes-attestation-guardian` and documented PR-claim-to-code/test traceability in human-friendly terms.
- 2026-04-15: Replaced table-style PR claim mapping with full narrative claim breakdowns (people-speak, wiring, verification, and concrete scenarios per claim).
## Source References
- skills/hermes-attestation-guardian/skill.json
@@ -61,4 +285,3 @@ node skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs
- skills/hermes-attestation-guardian/test/attestation_diff.test.mjs
- skills/hermes-attestation-guardian/test/attestation_cli.test.mjs
- skills/hermes-attestation-guardian/test/setup_attestation_cron.test.mjs
- docs/plans/2026-04-15-hermes-attestation-guardian-draft.md