agentleFS
svy04svy04/ballast/skills/pin/SKILL.md

pin

Skill73 starsChanged 3 months agoMITView on GitHub

Turn a correction or working preference the user just expressed into a permanent ballast rule. Use when the user corrects your behavior, says "don't do X" / "always do Y" / "I told you before", or asks you to remember how they want things done.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before writing the rule, name which net failed — the remedy differs by type:

    line 12
  2. Distinctive:

    Never pin silently: Always show the entry once before writing — the user is the author of their own rules.

    line 48
  3. Distinctive:

    Offer proactively: When the user corrects the same behavior a second time, or says "I told you", offer to pin it — that phrase is the exact …

    line 49

≈ “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

Pin — corrections become rules

A correction that lives only in the conversation dies with the conversation. Pinning it writes it into the ballast rule catalog, so the rules hook delivers it with every future message it applies to.

When the correction follows an incident, classify first

Before writing the rule, name which net failed — the remedy differs by type:

  • default regression — the rule existed but wasn't in front of the generating moment → fix delivery: keywords that will actually match future messages
  • propagation miss — one surface got fixed, its copies didn't → sweep the remaining surfaces now, then pin; when a ledger decision changed, decision-ledger's supersede sweep is the same move
  • delegation leak — the work went through a subagent or external tool the hook never reaches → the rule belongs in the brief, not only the catalog
  • variant evasion — only the literal wording was watched and a rephrasing walked through → pin patterns, not just keywords
  • compression loss — shortening or re-toning a text dropped something the text had already settled → diff the trimmed version against the one before it, part by part; a rule pinned here belongs in the editing brief, not the keyword catalog
  • substitute illusion — something adjacent to what the rule asked for got done, and that was scored as compliance → check that the artifact the rule names actually exists, rather than that the work felt covered

The last two are worth separating from the first: they happen with the rule already in view, so better delivery does not reduce them. One clause in the entry's parenthetical is enough ("propagation miss: fixed README, missed plugin.json"). An apology without a classification fixes nothing. These six types are one owner's working taxonomy, and it grew — it started at four. When an incident fits none of them, that's the next type worth naming.

Steps

  1. Extract the rule. One imperative sentence, at most two. The user's own wording beats your paraphrase. If the correction references a specific incident, keep the incident as a short parenthetical — future-you needs the why.
  2. Propose the entry in one compact block and ask nothing else:

    • id: short kebab-case
    • title: a few words — this is the label shown when the rule is injected
    • when.keywords: 4–8 keywords likely to appear in future messages where this rule matters (case-insensitive substring match) — in the language(s) the user actually types. Too-generic keywords ("please", "make") spam every turn; too-narrow ones never fire. For shapes keywords can't catch, add when.patterns (regex, case-insensitive). when.always: true fires on every message — reserve it for 1–2 rules at most.
    • body: the rule text
    • action: omit it (inject is the default) or "block" if the user wants matching requests stopped outright

    Example entry, exactly as it will sit in the catalog:

    {
      "id": "cost-gate",
      "title": "Estimate before spending",
      "when": { "keywords": ["generate", "credits", "batch"] },
      "body": "Anything that spends money or credits: present an estimate and get explicit approval BEFORE executing. No exceptions for small amounts — the habit is the point."
    }
  3. On the user's OK (a plain "yes/좋아/그래" is enough), merge it into <project>/.claude/ballast.rules.json — or ~/.claude/ballast.rules.json when the rule is about how the user works everywhere, not just this project. Create the file with {"version": 1, "rules": []} if missing; if an entry with the same id exists, update its body and say what changed instead of duplicating.
  4. Confirm in one line: "Pinned <id> — it now arrives with every matching message." Nothing more.

Rules

  • Never pin silently. Always show the entry once before writing — the user is the author of their own rules.
  • Offer proactively. When the user corrects the same behavior a second time, or says "I told you", offer to pin it — that phrase is the exact pain this skill exists for.
  • Keep catalogs lean. Past ~20 rules, suggest pruning before adding: an injected rule nobody reads is worse than none. The hook caps injection at 12 rules / ~6,000 characters.
  • User-level vs project-level. Rules about how the user works everywhere go to ~/.claude/ballast.rules.json; rules about this project stay in the project. Ask only when it's genuinely ambiguous.
  • Blocks are for danger, not style. Reserve action: "block" for irreversible or costly actions; style preferences are injections.

Part of ballast — foundations first, every solved path kept, done means checks pass.

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.