← Back to workflows
Live Operations Medium External bundle v1.0.0

Recursive Self-Improvement Pack

A recursive workflow with memory, guardrails, and teeth.

A workflow bundle for turning agent mistakes into durable system changes through daily reflection, candidate promotion, skill evolution, recidivism enforcement, and visibility reports.

Availability
Classification
External bundle
Install mode
Agent-installable
Review status
Source review
Verification mode
Manual verification

This workflow now publishes a real installable bundle at a stable SuperAda URL plus a reviewable GitHub source path. It installs workflow assets into an OpenClaw workspace, then requires verification before any cron activation.

Install source
SuperAda install bundle
bash <(curl -fsSL https://superada.ai/install/recursive-self-improvement-pack/install.sh) /path/to/your/openclaw-workspace
Workflow record
Bundle id
superada.workflow.recursive-self-improvement-pack
Entrypoint
install.sh
Bundle root
https://superada.ai/install/recursive-self-improvement-pack/
Artifact count
9 described items

Daily reflection, promotion, skill evolution, enforcement, and visibility artifacts that work together as one recursive improvement loop.

Install contract
Installability
Truly installable now
Method
Installable via script
Install source spec
https://superada.ai/install/recursive-self-improvement-pack/install.sh
bash <(curl -fsSL https://superada.ai/install/recursive-self-improvement-pack/install.sh) /path/to/your/openclaw-workspace

Canonical source: https://superada.ai/install/recursive-self-improvement-pack/

Agent step 1
Review the bundle README, install manifest, and GitHub source path before running anything.
Agent step 2
Run `bash <(curl -fsSL https://superada.ai/install/recursive-self-improvement-pack/install.sh) /path/to/your/openclaw-workspace`.
Agent step 3
Review thresholds, memory mutation behavior, and installed docs before enabling any cron.
Agent step 4
Run the quick verification commands in the bundle README before trusting the automated path.
Verification command
bash scripts/crons/daily-soul-review.sh && bash scripts/crons/recursive-improvement.sh && bash scripts/crons/weekly-self-improvement-review.sh && python3 scripts/crons/generate-recursive-visibility.py
Current limits
  • The installer copies assets into a workspace but intentionally does not auto-create cron jobs.
  • The installer does not auto-enable automation or mutate live config outside the installed files.
  • Operators still need to review thresholds and memory mutation behavior before activation.
Step 1
Review the published install bundle and source
Inspect the README, install script, bundle manifest, and GitHub source path before running anything.
Step 2
Run the installer into your workspace
Use the published installer to copy the bundle into the target OpenClaw workspace.
bash <(curl -fsSL https://superada.ai/install/recursive-self-improvement-pack/install.sh) /path/to/your/openclaw-workspace
Step 3
Review thresholds and memory mutation behavior
Check recidivism thresholds, rule promotion behavior, and any direct MEMORY.md insertion before enabling automation.
Step 4
Wire the cron layer
Recreate the daily soul review and recursive improvement cron triggers in the target workspace.
Step 5
Run a quick verification pass
Generate one daily report, one recursive promotion pass, one weekly review, and one visibility report before enabling unattended use.

If you hand an agent this workflow URL, this install contract is the source of truth for whether it can really install the bundle now, what source to use, and what still needs a human.

What you get

  • Restore a dedicated daily reflection loop
  • Promote recurring failures into rules, skills, or workflow patches
  • Enforce guardrails when recidivism crosses thresholds
  • Generate visibility reports for the recursive funnel

Workflow includes

  • Daily soul review loop
  • Recursive improvement pipeline
  • Skill evolution review
  • Recidivism enforcement
  • Recursive visibility reports

Described artifacts

Installer script
script
https://superada.ai/install/recursive-self-improvement-pack/install.sh
Copies the bundle assets into a target OpenClaw workspace.
Bundle README
doc
https://superada.ai/install/recursive-self-improvement-pack/README.md
Published install instructions, verification steps, and operator warnings.
Daily soul review script
script
https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/daily-soul-review.sh
Restores the explicit daily introspection loop and generates a daily self-improvement report.
Recursive improvement pipeline spec
doc
https://superada.ai/install/recursive-self-improvement-pack/docs/specs/recursive-improvement-pipeline.md
Canonical capture -> classify -> promote -> apply -> verify workflow spec.
Improvement candidate template
template
https://superada.ai/install/recursive-self-improvement-pack/docs/templates/improvement-candidate.md
Standardized candidate structure used across the recursive pipeline.
Recursive improvement cron script
script
https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/recursive-improvement.sh
Promotes high-signal reflections into workflow candidates and triggers recidivism enforcement.
Skill evolution path spec
doc
https://superada.ai/install/recursive-self-improvement-pack/docs/specs/skill-evolution-path.md
Explains how repeated failures turn into skill patches, helper scripts, or new skills.
Skill evolution review script
script
https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/skill-evolution-review.sh
Reviews skill-related candidates and separates safe auto-apply from human-gated changes.
Weekly self-improvement review script
script
https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/weekly-self-improvement-review.sh
Summarizes proposed and applied candidates and checks which ones need verification.
Recidivism enforcement policy and script
script
https://superada.ai/install/recursive-self-improvement-pack/scripts/enforce-recidivism.py
Promotes repeated failures into enforced guardrails and escalates persistent recidivism into structural candidates.
Recursive visibility spec and generator
script
https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/generate-recursive-visibility.py
Generates visibility outputs for funnel states, promoted rules, recidivism, and bottlenecks.

Requirements

OpenClaw workspace with cron support
dependency
The target environment needs a working OpenClaw install and an editable workspace.
Memory files and reflections log
dependency
The workflow expects daily memory notes, a reflections log, and a recidivism tracker.
Human review before production use
review
The bundle is intentionally review-first because it can promote rules and mutate memory state.

Declared structure

  • https://superada.ai/install/recursive-self-improvement-pack/
  • https://superada.ai/install/recursive-self-improvement-pack/install.sh
  • https://superada.ai/install/recursive-self-improvement-pack/README.md
  • https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/daily-soul-review.sh
  • https://superada.ai/install/recursive-self-improvement-pack/docs/specs/recursive-improvement-pipeline.md
  • https://superada.ai/install/recursive-self-improvement-pack/docs/templates/improvement-candidate.md
  • https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/recursive-improvement.sh
  • https://superada.ai/install/recursive-self-improvement-pack/docs/specs/skill-evolution-path.md
  • https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/skill-evolution-review.sh
  • https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/weekly-self-improvement-review.sh
  • https://superada.ai/install/recursive-self-improvement-pack/scripts/enforce-recidivism.py
  • https://superada.ai/install/recursive-self-improvement-pack/scripts/crons/generate-recursive-visibility.py

Verification

Verification language now matches reality. Installable bundles can be tested after install, external bundles should be source-reviewed first, and conceptual entries are reviewed as designs until real artifacts exist.

Daily introspection loop runs cleanly
Execute the daily soul review script as a quick post-install sanity check.
bash <(curl -fsSL https://superada.ai/install/recursive-self-improvement-pack/install.sh) /path/to/your/openclaw-workspace && bash scripts/crons/daily-soul-review.sh
Expected: A daily report appears under output/self-improvement/.
Recursive promotion path works
Run the recursive improvement script after adding or reusing a high-signal reflection.
bash scripts/crons/recursive-improvement.sh
Expected: Candidate files are promoted and visibility generation runs.
Weekly review summarizes the queue
Run the weekly review script manually.
bash scripts/crons/weekly-self-improvement-review.sh
Expected: A weekly report appears with candidate funnel states and verification reminders.
  • Confirm the daily soul review generates an output file.
  • Confirm the recursive pipeline promotes candidates without duplicating prior ones.
  • Confirm recidivism enforcement only promotes the intended guardrails.

Best used for

  • Agent ops teams that keep repeating the same mistakes
  • Builders who need a promotion path from reflection to structural change
  • Operators who want self-improvement with proof instead of vibes

Operator notes

  • This is a workflow, not a single skill. The value comes from the chained loop.