Skip to content

About

Standard-library Python utilities for file hashing, artifact metadata, lightweight contract validation, and persistent change checks, with historical automation demo notes.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

AgentForge — Artifact Contracts & Change Detection

Source guide to AgentForge artifact helpers, validation rules, and demo records

Source guide drawn from the files in this repository; not a runtime screenshot or a fresh benchmark.

Small Python utilities for checking artifacts and detecting when inputs change. AgentForge contains file hashing, a persistent hash store, artifact metadata, and lightweight validators. These are building blocks for automation workflows, not an installable multi-agent framework or an LLM runtime.

What is here

Capability Source-backed behavior
File hashing Stream a file through SHA-256, SHA-1, or MD5; optionally persist the digest.
Change checks Compare a caller-supplied hash with a JSON store and report NEW, UNCHANGED, or CHANGED.
Artifact management Write UTF-8 text plus a .meta.json sidecar; read or recursively list files.
Contract validation Check text directives, a minimal JSON key/type contract, or CSV columns and row bounds.
Monitor example Illustrate hash-based no-op detection using an installation-specific script.

The utilities use Python's standard library. There is no package manifest, requirements.txt, or agentforge.Forge implementation in this repository.

Getting started

Use Python 3 and run the scripts directly from the repository root:

git clone https://github.com/MdSadman2004/AgentForge.git
cd AgentForge
python src/artifact_hash.py README.md
python src/artifact_manager.py list --dir src
python src/validate_artifact.py --help

Hashing defaults to SHA-256. MD5 and SHA-1 are compatibility options, not recommendations for security-sensitive integrity checks.

Write and inspect an artifact

These commands create a demo file and its metadata sidecar:

python src/artifact_manager.py write --path demo-output/note.txt --content "Example artifact"
python src/artifact_manager.py read --path demo-output/note.txt

write replaces an existing artifact at the supplied path. Use a disposable output directory when experimenting.

Check without changing the store

The following uses an illustrative caller-supplied token, not a file digest:

python src/change_detector.py --store demo-output/hashes.json --key example --hash demo-v1
python src/change_detector.py --store demo-output/hashes.json --key example --hash demo-v1 --action store
python src/change_detector.py --store demo-output/hashes.json --key example --hash demo-v1

The first check reports NEW; storing is a separate action. For real change detection, pass the digest emitted by artifact_hash.py. The parser uses --store and --action; it has no --check argument.

Source guide

Start here Purpose
Hash utility Chunked file hashing and optional digest output.
Change detector JSON hash store and explicit check/store actions.
Artifact manager Text files, sidecar metadata, and recursive listing.
Validator Contract formats, supported directives, and exit codes.
Monitor demo Installation-specific no-op monitor example.
Historical verification note Notes about a wider local Hermes setup, not bundled skills.
Demo analysis report Recorded example comparison; not a reproducible benchmark suite.

Scope & limitations

  • No planner, agent classes, vector-memory service, or model integration is bundled.
  • The JSON validator implements its own limited contract, not the JSON Schema standard. YAML extensions select text-rule validation; there is no YAML parser.
  • CSV row bounds count physical lines, so multiline CSV fields need extra care.
  • Hash-store writes are not locked or transactional; concurrent writers can race.
  • monitor_demo.py hardcodes Windows paths outside the checkout. Adapt those paths before use. Its implemented change branch exits 0, despite its header describing exit code 2; it does not check subprocess failures.
  • The demo scores and SYSTEM_VERIFIED.md describe historical artifacts and a broader local installation. The underlying datasets, evaluation generators, and referenced Hermes skills are not included here.
  • No tests or benchmark reproduction were run for this documentation refresh.

Credits and license

Original README notice: MIT © Md Sadman Bin Masud. No standalone license file is present. This refresh preserves the existing notice without adding license terms or changing the project's licensing.

About

Standard-library Python utilities for file hashing, artifact metadata, lightweight contract validation, and persistent change checks, with historical automation demo notes.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages