Dream rollback
Skill vickysrawat/AI-Assisted-development/skills/dream-rollback
Reverses a specific Dream consolidation run using the audit trail in memory/dream-log.md. Use when a dream run promoted incorrect information, wrongly deleted an important memory, or made changes that need reversing. Triggers on: "rollback dream", "undo dream", "reverse dream run", "dream ran incorrectly", "restore memory", "undo last dream", or any request to revert Dream consolidation changes.From its SKILL.md
npx -y skills add vickysrawat/AI-Assisted-development --skill dream-rollbackAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
7.9 KB, as published. Nobody here has run it
Dream Rollback Skill
Skill version: 1.0 · Last changed: 2026-06-03 · Plugin compatibility: ≥1.14.0 · Consent: C
Reads memory/dream-log.md to identify and reverse a specific dream run.
The rollback itself is logged as a new entry — making it auditable and itself reversible.
Step 1 — Read the dream log
cat memory/dream-log.md 2>/dev/null || echo "NO_LOG"
If NO_LOG → output:
⚠ No dream-log.md found. Dream rollback requires an existing audit log.
Run /dream first to create one, then use rollback if needed.
And stop.
Step 2 — List available runs
Parse memory/dream-log.md for run entries. Each run entry has:
- A run ID (date + sequence number, e.g.
DREAM-2026-05-30-1) - A date
- A summary of changes (ADD/UPDATE/DELETE counts)
- The specific operations performed
Display the last 10 runs in reverse chronological order:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Dream Run History (last 10)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
[1] DREAM-2026-05-30-2 30 May 2026 3 ADD · 1 UPDATE · 0 DELETE
[2] DREAM-2026-05-30-1 30 May 2026 0 ADD · 2 UPDATE · 1 DELETE
[3] DREAM-2026-05-24-1 24 May 2026 5 ADD · 0 UPDATE · 0 DELETE
...
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Which run to roll back? (Enter number or run ID, or 'cancel')
Wait for developer input.
Step 3 — Show changes in the selected run
Once the developer selects a run, display every operation from that run:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
DREAM-2026-05-30-2 — Changes to reverse
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
ADD → CLAUDE.md
Topic: "Always use EF Core for database access"
Added: lines 47-49 in # Dream → promoted facts
ADD → memory/topic-dotnet.md
Section: "Repository pattern conventions"
UPDATE → CLAUDE.md
Topic: "Branch naming convention"
Before: "feature/ADO-[ID]-description"
After: "feat/ADO-[ID]-description"
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Reverse all 3 operations? (yes / partial / cancel)
'partial' lets you choose which operations to reverse individually
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Wait for developer input.
Step 4 — Execute rollback
For each operation to reverse (based on developer selection):
Reversing an ADD:
- Remove the added content from the target file
- If an entire section was added, remove the section
- If lines were appended, remove those lines
- If a file was created by this run, delete it
Reversing an UPDATE:
- Restore the "Before" content in place of the "After" content
- Use the exact before-state recorded in the dream log
Reversing a DELETE:
- Restore the deleted content to its original location
- If a file was deleted, recreate it with the original content
After each operation, confirm to the developer:
✓ Reversed ADD: "Always use EF Core..." removed from CLAUDE.md
✓ Reversed UPDATE: Branch naming restored to "feature/ADO-[ID]-description"
✓ Reversed ADD: memory/topic-dotnet.md section removed
Step 5 — Log the rollback
Append a rollback entry to memory/dream-log.md:
---
## ROLLBACK — {date}
**Rolled back**: {run ID}
**Reason**: {developer's stated reason if provided, else "not stated"}
**Operations reversed**: {N}
### Operations
- ↩ ADD reversed: "{topic}" removed from {file}
- ↩ UPDATE reversed: "{topic}" in {file} — restored to before-state
- ↩ ADD reversed: {file} section removed
**Note**: This rollback can itself be rolled back by referencing this log entry.
Step 6 — Confirm
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✅ Rollback complete
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Run reversed : {run ID}
Operations : {N} reversed
Log entry : ROLLBACK-{date} appended to memory/dream-log.md
This rollback is itself auditable and reversible.
Run /dream-health to see the updated memory state.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Model routing
This skill is in the infrastructure tier — it uses INFRA_MODEL
(default: claude-sonnet-4-6).
memory revert — reads audit trail and reverses consolidation runs
To override for this project:
{ "env": { "INFRA_MODEL": "claude-opus-4-8" } }
See $PLUGIN_DIR/skills/shared/model-routing-spec.md for the full routing specification.
Persona
Acts with a [DPE] DevOps/Platform Engineer lens — state integrity, safe reversal, auditability;
always asks "what happens on partial failure, and is this reversible?" Lens only; never assume, never
attribute in output. See $PLUGIN_DIR/skills/shared/personas-spec.md.
Business context severity
This skill does not perform security or compliance reviews. If output from this
skill surfaces data that may trigger B1–B7 sensitivity (see
$PLUGIN_DIR/skills/shared/business-context-severity.md), flag it to the developer. Do not
silently process or display attorney-client privileged matter data, immigration
identifiers, or other B1–B7 categories without acknowledgement.
Hard Rules
- NEVER rollback without explicit developer confirmation at Step 4
- NEVER delete a topic file if it contains content from multiple dream runs — only remove the sections added by the specific run being reversed
- NEVER modify dream-log.md entries — only append the rollback record
- NEVER roll back ledger files (
CodeReviews/,security/,dynamic-scan/) — these track real defects across runs and are managed by their respective skills and/fix. Rolling back a ledger would hide known vulnerabilities. Use the skill's own re-scan to refresh it, or/fixto resolve individual findings - NEVER roll back
.claude/architecture/files — use/update-archinstead - If a "Before" state is not recorded in the log for an UPDATE, warn the developer and skip that operation rather than guessing
- If the target file has changed since the dream run (different content than
the "After" state), do not silently proceed. Show a diff between the log's
"After" state and the current file content so the developer can see exactly
what would be overwritten, then offer three options:
Wait for explicit choice before touching the file.⚠️ {file} has been modified since this dream run. Changes since the run: - Line 14: "{after-state-line}" → "{current-line}" - Line 22: "{after-state-line}" → "{current-line}" Options: proceed — overwrite with Before state (manual edits since the run will be lost) skip — leave this file unchanged, continue with remaining operations cancel — abort the entire rollback - Partial rollbacks are valid — the developer may choose to reverse only specific operations from a run