agentleFS
Sign inStart free
appleweipingappleweiping/AGENT_RESOURCE/skills/vipin/project-hardness/SKILL.md

project-hardness

Skill78 starsChanged 24 days agoApache-2.0View on GitHub

Turn any local project into an AI-readable, maintainable, verifiable causal layer under the project's `.agent/` directory, and harden vague requirements into concrete task specs before editing. Use when (a) starting work on an unfamiliar or large project, (b) you need to know a change's blast radius / do-not-modify boundaries, or (c) a requirement is vague and you need clarified goal + involved modules + edit paths + risks + acceptance criteria + test commands. Not a grep/index/chat-memory; it is a static, source-backed project model. Applies to every agent (Claude Code, Codex, OpenCode, etc.).

  • Reads credentials

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Only cross-project lessons are ever synced to agentmemory — project facts stay local.

    line 16
  2. Very specific:

    After completing work → update .agent/ (re-scan) and sync only reusable lessons.

    line 23
  3. Very specific:

    Run from the root of a WEIPING_WIKI checkout:

    line 27
  4. Very specific:

    Before editing: check impact / the module's blast radius.

    line 70

≈ “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

project-hardness

A per-project causal layer: a static, source-backed model of a project written to <project>/.agent/. It answers "what does this project do, what depends on what, what breaks if I change X, what must I not touch, and why was it built this way" — without re-reading the whole repo every session.

The engine lives in WEIPING_WIKI (scripts/hardness/, invoked via python scripts/wiki.py hardness). The artifacts live in the target project (.agent/, git-ignored by default). Only cross-project lessons are ever synced to agentmemory — project facts stay local.

When to use

  • Starting work on an unfamiliar, large, or long-dormant project → scan first.
  • About to change something and unsure of impact → impact <module> or read modules/<m>.md.
  • A requirement is vague → harden it into a task spec BEFORE writing code.
  • After completing work → update .agent/ (re-scan) and sync only reusable lessons.

Commands

Run from the root of a WEIPING_WIKI checkout:

# 1. Build / refresh the causal layer for a project
python scripts/wiki.py hardness scan <project-path> [--name NAME] [--json]

# 1b. Discover and scan every important project on the machine (de-duped, safety-excluded)
python scripts/wiki.py hardness scan-all [--dry-run] [--json]

# 2. Harden a vague requirement into a concrete task spec (.agent/tasks/<id>.md)
python scripts/wiki.py hardness harden <project-path> "the requirement text" [--id ID]

# 3. Blast radius of a single module
python scripts/wiki.py hardness impact <project-path> <module-name>

# 4. Sync ONLY cross-project lessons to agentmemory (project facts are auto-rejected)
python scripts/wiki.py hardness sync <project-path> --lesson "general reusable lesson" [--commit]

sync is dry-run by default; pass --commit to actually send. It refuses any text that names concrete files, paths, or line numbers (those are project facts, not lessons).

What .agent/ contains

ArtifactPurpose
architecture.mdModule table, responsibilities, dependency hotspots
glossary.mdDomain terms, entities, frameworks
constraints.mdDo-not-modify boundaries (secrets, lockfiles, migrations, markers)
decisions.mdDecision-flavored git history (manual additions preserved on regen)
modules/*.mdPer-module detail + blast radius + files
flows/*.flow.mdBusiness/process flows (route → handler → deps → entities)
entities.jsonEntities and their lifecycle / touching modules
causal-graph.jsonFull node/edge graph (imports, routes-to, reads/writes)
tasks/*.mdHardened task specs
index.jsonManifest + provenance

Standard workflow

  1. First contact: scan the project. Read architecture.md and constraints.md.
  2. Per requirement: harden it. Read the generated tasks/<id>.md — it lists the clarified goal, involved modules, likely edit paths, risk points, acceptance criteria, and test commands. Refine it if the match confidence is LOW.
  3. Before editing: check impact / the module's blast radius; never touch block constraints without explicit user approval.
  4. After completing: re-scan to refresh .agent/; run the acceptance test commands; then sync only generalizable lessons to agentmemory.

Boundary rules (hard)

  • Project-specific facts stay in .agent/. Never write them to global agentmemory.
  • The engine never reads secret files (.env, *.key, credentials.*) — it records their existence as a block constraint and moves on.
  • .agent/ is git-ignored by default (it is a derived artifact, regenerable any time).
  • The engine is stdlib-only and fails gracefully if agentmemory is unreachable.

See the full design at wiki/concepts/project-hardness-system.md in WEIPING_WIKI.

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.

Check your setup, free