agentleFS
iwesysiwesys/IWE/.claude/skills/run-protocol/SKILL.md

run-protocol

Skill0 starsChanged 24 days agoMITView on GitHub

Step-by-step execution of the OWC protocol with mandatory checkpoint

3 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check and read (if they exist):

    line 35
  2. Distinctive:

    Does not exist → skip silently.

    line 40
  3. Distinctive:

    BLOCKING: Do NOT skip Steps. If a Step is impossible — mark it as blocked and ask the user.

    line 65
  4. Distinctive:

    After all closing Steps are complete:

    line 72
  5. Very specific:

    Only one Step in_progress at a time

    line 79
  6. Distinctive:

    Do not run ahead — context becomes contaminated (SOTA.002)

    line 80
  7. Distinctive:

    If the Protocol is interrupted by the user — write the remaining Steps to a checkpoint

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Protocol Execution

Principle: A Protocol is a sequence of Steps. Mark each Step complete before moving to the next. Skipping Steps is not allowed. Problem solved: The agent "forgets" final Steps (Verification, Backup) due to context contamination (SOTA.002).

Arguments: $ARGUMENTS

Step 1. Identify the Protocol

ArgumentRouting (short)Full algorithm
day-open / open daymemory/protocol-open.md (§ Scope: Day).claude/skills/day-open/SKILL.md
open session or taskmemory/protocol-open.md (§ Scope: Session)— (inline in protocol-open.md)
day-close / close daymemory/protocol-close.md (§ Routing).claude/skills/day-close/SKILL.md
close sessionmemory/protocol-close.md (§ Quick Close)— (inline in protocol-close.md)
week-closememory/protocol-close.md (§ Routing).claude/skills/week-close/SKILL.md

If a Skill file exists for the argument → read it (contains the full algorithm and checklist). The protocol file is a slim router + Quick Close inline.

Step 1b. Load extensions (BLOCKING)

Identify the protocol name: day-open, day-close, week-close, protocol-close, protocol-open.

Check and read (if they exist):

  1. extensions/{protocol}.before.md → add as the first Steps in TodoWrite
  2. extensions/{protocol}.after.md → add as Steps after the main algorithm, before Verification
  3. extensions/{protocol}.checks.md → add as a Step before git commit (BLOCKING: commit is forbidden until checks pass)

Does not exist → skip silently. Exists → read and include in the plan.

Step 2. Extract Steps

From the protocol algorithm (Skill file or protocol file) and extensions, extract a numbered list of Steps. Record them as tasks (TodoWrite; the tool may be unavailable in the session — on models newer than Sonnet 4.6 it is disabled by the client by default, enabled via the CLAUDE_CODE_ENABLE_TODO_TOOLS=1 variable in the env settings (issue #1216); until then, or on installations without this variable, track the same Steps using explicit numbering in responses, marking each one complete before moving to the next, and notify the pilot in one line — issues #561, #563):

Task order:

  1. Extensions .before.md (if present)
  2. Main Steps from the algorithm
  3. Extensions .after.md (if present)
  4. Extensions .checks.md + git commit (if there is an Artifact to commit)
  5. Closing Verification (sub-agent Haiku R23 against the closing checklist) — closing Protocols only, see Step 4
  • Each main Step = a separate task
  • For closing Protocols, the last task = "Closing Verification (sub-agent Haiku R23)"; opening and other Protocols do not include this task
  • Status: pending

Step 3. Execute Sequentially

For each Step:

  1. Mark the Step as in_progress
  2. Execute it
  3. Mark it as completed
  4. Move to the next Step

BLOCKING: Do NOT skip Steps. If a Step is impossible — mark it as blocked and ask the user.

Step 4. Closing Verification (closing Protocols only)

This Step applies only during closing: Quick Close (close, close session), Day Close, Week Close, Month Close. Opening and other Protocols do not include R23 Verification; skip this Step. Quick Close: if params.yaml has verify_quick_close: false — skip this Step. Day Close, Week Close, and Month Close are not subject to this key.

After all closing Steps are complete:

  1. Launch sub-agent Haiku in the R23 role (context isolation): pass the closing checklist (from the protocol or closing Skill), the list of modified files, and a brief summary
  2. R23 compares execution against the closing checklist
  3. For each ❌ — fix the issue BEFORE showing the result to the user

Rules

  • Only one Step in_progress at a time
  • Do not run ahead — context becomes contaminated (SOTA.002)
  • Record Protocol Steps as work progresses, not at the moment of compression: the agent receives nothing on PreCompact. The precompact-checkpoint.sh Hook takes a snapshot of the conversation turns (outside the project, at $CLAUDE_CONFIG_DIR/iwe-checkpoints/); precompact-restore.sh restores it after compression — but only what has already appeared in the most recent turns, not the Protocol Step separately
  • If the Protocol is interrupted by the user — write the remaining Steps to a checkpoint

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.