mcpField
L
learning-loop

learning-loop

github.com/richfremAgent workflows738 installsUpdated 1d ago
Repository

About learning-loop

A cognitive continuity protocol for carrying knowledge across isolated agent sessions. It requires session context before code changes and structures work through research, human approval, adversarial review, execution, and closure.

Key features

  1. 01Carry knowledge across sessions
  2. 02Organize research and execution into phases
  3. 03Require explicit human approval before proceeding
  4. 04Add red-team review before execution
  5. 05Record a retrospective at completion

Use cases

  1. 01Maintain project context across isolated sessions
  2. 02Record research findings in memory or a wiki
  3. 03Add approval and adversarial review before implementation
  4. 04Complete work through a defined process and log a retrospective

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/richfrem/agent-plugins-skills --skill learning-loop

After installing, type/learning-loopin Claude Code to confirm it loaded.

Bundled filesevals/evals.jsonevals/results.tsv

Details

From the project README

Dependencies

Requires Python 3.8+ (standard library only).

Learning Loop (learning-loop)

Cognitive continuity protocol ensuring knowledge survives across isolated agent sessions.

Contents

Constraints

  • Anti-simulation rule: Never describe what you "would do" or mark steps complete without executing them.
  • Mandatory closure: Full closure sequence must be performed upon session completion.
  • Prerequisite context: Always establish valid session context upon Wakeup before modifying code.

Quick start

# Initialize learning loop session and load orientation context
python3 -c "import sys; print('Learning loop initialized across Orientation -> Synthesis -> Gate')"

Workflow

Orientation -> Synthesis -> Strategic Gate -> Red Team Audit -> [Execution] -> Completion
  1. Phase I (Orientation): Read local primers, load session state, verify readiness.
  2. Phase II (Synthesis): Conduct research and record modular findings in memory/wiki.
  3. Phase III (Strategic Gate): Present findings to human; require explicit approval to proceed.
  4. Phase IV (Red Team Audit): Submit research packet to adversarial review before execution.
  5. Phase V (Execution & Completion): Single Loop (solo implementation) or Dual Loop (delegation); log retrospective.

Verification

# Verify session completion and check retrospective entry
test -f references/phases.md && echo "Phase contracts valid"
git status --short

References

Comments

Markdown supported