Skip to content

identity-continuity.md

Latest commit

 

History

History
129 lines (96 loc) · 3.26 KB

identity-continuity.md

File metadata and controls

129 lines (96 loc) · 3.26 KB
Raw

Identity Continuity

Identity continuity adds recovery-oriented memory artifacts so the assistant can regain stable behavior after drift, context loss, or tool/runtime incidents.

This feature set is controlled by the v8.4 config surface and is designed to be fail-open: disabling continuity flags should return runtime behavior to baseline retrieval/extraction paths.

Artifacts

When enabled, continuity files are stored under:

<memoryDir>/identity/

Primary artifacts:

  • identity-anchor.md: canonical continuity anchor sections.
  • incidents/*.md: incident records with open/close lifecycle.
  • audits/weekly/*.md and audits/monthly/*.md: generated continuity audits by period.
  • improvement-loops.md: recurring loop register and review metadata.

Safety Boundaries

Continuity features must keep these invariants:

  1. No mutation of OpenClaw session pointers/files.
  2. Incident lifecycle is append-only except explicit close transition.
  3. Identity injection respects identityInjectionMode and identityMaxInjectChars.
  4. Disabled flags are compatibility guarantees, not hints:
    • identityContinuityEnabled=false disables continuity injection/tools.
    • continuityIncidentLoggingEnabled=false disables incident logging paths.
    • continuityAuditEnabled=false disables audit generation paths.
  5. Fail-open behavior on parse/storage errors (log and continue).

Template: Identity Anchor

Use this structure for safe merges via identity_anchor_update:

# Identity Anchor

## Identity Traits
- Role:
- Core strengths:
- Reliability profile:

## Communication Preferences
- Tone:
- Detail level:
- Avoid:

## Operating Principles
- Principle 1:
- Principle 2:

## Continuity Notes
- Active risks:
- Recent corrections:
- Recovery guidance:

Template: Continuity Incident

Incident files are markdown with frontmatter; open/close tools maintain lifecycle fields.

---
id: incident-<ts>-<slug>
state: open
openedAt: 2026-02-25T00:00:00.000Z
closedAt:
---

## Timeline
- 2026-02-25T00:00:00.000Z opened

## Symptom
identity anchor omitted in recovery response

## Fix Applied

## Verification Result

## Notes
Observed during weekly continuity audit.

Template: Continuity Audit

---
id: continuity-audit-2026-02-25
period: weekly
generatedAt: 2026-02-25T00:00:00.000Z
signalSummary:
  openIncidents: 1
  staleLoops: 2
  anchorPresent: true
---

# Continuity Audit

## Signal Checks
- Anchor present: pass
- Incident backlog: warn
- Improvement-loop freshness: warn

## Findings
- Incident `incident-...` still open past target SLA.
- Two active loops exceeded cadence threshold.

## Recommended Actions
- Close incident after verification.
- Run `continuity_loop_review` for stale loops.

Rollout by Risk Tier

  1. Low risk:
    • Enable identityContinuityEnabled.
    • Keep identityInjectionMode=recovery_only.
    • Leave continuityIncidentLoggingEnabled and continuityAuditEnabled off.
  2. Medium risk:
    • Enable incident logging and weekly audits.
    • Keep explicit alerting on stale loops and open incidents.
  3. High risk:
    • Enable full continuity workflow with regular audit cadence.
    • Add operator review gate before any mode shift to full.
    • Require hardening checks before merge for continuity-path changes.