- Home
- Skills
- Agents & Orchestration
- Multi-Agent Delivery & Hygiene Pack
Works with the AI Agents you already use
Multi-Agent Delivery & Hygiene Pack
This pack gives multi-agent work a paper trail: who made and who verified, exactly what was reviewed, what happened in the log, and what to cut from memory.
$29
Multi-Agent Delivery & Hygiene Pack
Example session with this skill installed
Verify Maker/Verifier separation in bad.json, generate a SHA-256 deliverable manifest for out/, extract verbatim errors/fixes from session.jsonl, and audit memory file bloat.
- Read your context and instructions
- Compiled the multi-agent delivery
Real output from running the packaged scripts on small fixtures.
1. node contract_gate_runner.mjs separation bad.json
$ node contract_gate_runner.mjs separation bad.json
contract_gate_runner separation: FAIL
[error] same-agent - Maker and Verifier have the same id (agent-a).
[error] same-session - Maker and Verifier share one session; the Verifier sees the Maker's reasoning and inherits its blind spots.
[warn] same-model-family - Same model family (model); prefer a different family or at least a fresh context.
[error] verifier-can-edit - The Verifier can edit the deliverables; it must only read and report.
[error] no-verifier-evidence - The Verifier listed no evidence (commands run, exit codes, line numbers).
[error] no-deliverables - The contract names no deliverables.
(exit 1)
2. node contract_gate_runner.mjs manifest out
$ node contract_gate_runner.mjs manifest out
b6a98d9ce9a2d9149288fa3df42d377c3e42737afdcdaf714e33c0a100b51060 6 a.txt
f2c82decdd7181cf98945929a62598db7e6b477e11f6e0eb0ae97020eff151ad 5 sub/b.txt
contract_gate_runner manifest: PASS
(exit 0)
3. node session_log_distiller.mjs session.jsonl
$ node session_log_distiller.mjs session.jsonl
# Objective packet
Source: session.jsonl (3 records read). Method: verbatim line extraction by pattern, no summary.
## commands (1)
- [line 1] Bash: npm test
## errors (2)
- [line 2] Error: ENOENT no such file
- [line 2] exit code 1
## fixes (2)
- [line 3] Edit: src/a.js
- [line 3] Fixed it. This should work, probably. [U]
## unverified (1)
- [line 3] Fixed it. This should work, probably. [U]
(exit 0)
4. node memory_bloat_auditor.mjs bad --today 2026-11-17
$ node memory_bloat_auditor.mjs bad --today 2026-11-17
memory_bloat_auditor: 2 file(s), 292 characters
bad/CLAUDE.md (instructions, 193 chars)
[error] line 3 broken-link - Link target does not exist: docs/missing.md
fix: Fix the path or remove the link.
[error] line 3 broken-link - Wiki link [[nopage]] has no matching page in the scanned set.
fix: Create the page or fix the name.
[warn] line 7 stale-date - A "current state" line cites a date 676 days old.
fix: Re-verify the statement and update the date, or delete the line.
[warn] line 9 duplicate-heading - "Status" repeats the heading on line 5.
fix: Merge the sections or rename one.
bad/memory/a.md (note, 99 chars)
[warn] line 3 stale-date - A "current state" line cites a date 676 days old.
fix: Re-verify the statement and update the date, or delete the line.
[warn] line 3 duplicate-line - Same line as bad/CLAUDE.md:7: "currently on the old plan as of 2025-01-10, this line is lon"
fix: Keep it in one place and link to it.
summary: 2 error(s), 4 warning(s)
(exit 1)
Connects securely to your tools. The creator never sees your data.
About this skill
Stop taking the agent's word for it
An agent that checks its own work mostly agrees with itself, a summary of a long session carries the summarizer's blind spots, and instruction files only ever grow. This pack gives multi-agent work a paper trail: who made and who verified, exactly what was reviewed, what happened in the log, and what to cut from memory.
Each skill ships one self-contained Node.js script (Node 18+, standard library only). It prints findings or results with a fix, and exits non-zero when the gate fails. Add --json for machine-readable output.
The 3 skills
1. maker-verifier-delivery-loop enforces an independent Maker / Verifier loop:
- separation check: different id and session, Verifier is read-only and lists evidence
- SHA-256 manifest of the deliverables, then verify that nothing changed after review
- a gate that fails while
[U]unverified markers remain
2. unbiased-second-opinion-distiller turns a session log into an objective packet:
- reads JSONL, Markdown or text logs
- extracts commands, errors, fixes and unverified statements verbatim, with positions
- no summary and no interpretation, so the second reviewer starts from facts
3. agent-memory-and-wiki-hygiene audits CLAUDE.md, AGENTS.md, memory notes and wikis:
- character budgets per file and in total
- broken relative and wiki links, duplicate headings and repeated lines
- dated "current state" lines that went stale
Built for agents and for review
- Single file per skill, standard library only: copy one script anywhere and run it
- Works when the skill is loaded through the Agensi MCP: write the script to a temporary file, run it, delete it
- No network access, no shell commands, no environment reads
- Exit code and counts you can quote as evidence
What is in the ZIP
- README.md
- shared/verification-principles.md
- skills/agent-memory-and-wiki-hygiene/scripts/memory_bloat_auditor.mjs
- skills/agent-memory-and-wiki-hygiene/SKILL.md
- skills/maker-verifier-delivery-loop/scripts/contract_gate_runner.mjs
- skills/maker-verifier-delivery-loop/SKILL.md
- skills/unbiased-second-opinion-distiller/scripts/session_log_distiller.mjs
- skills/unbiased-second-opinion-distiller/SKILL.md
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Independence is shown by the fields you provide in the contract; the tool cannot see what a session contained.
Hashes prove files did not change, not that they are correct.
The distiller finds lines by pattern, not meaning: it can miss a failure phrased without the usual words.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed today
- Software
Node.js 18+
Local CLI tools only; only contract_gate_runner.mjs manifest --out writes the explicit JSON manifest path you pass.