folding-into-the-model
Skill2 starsChanged 4 days agoApache-2.0View on GitHub
Use when putting new information into the Keel model, or deciding whether it is already there - writing or changing a threat, weakness, mitigation or implementation, and working through incoming external material (an OWASP or ATLAS update, a CVE, an advisory, a whole taxonomy) to decide what the model already answers and what is a gap.
In this repository13 other files it gives its agents.
- assess-genai-with-librarySkill
- assessing-genai-securitySkill
- check-styleSkill
- humanizerSkill
- keel-designSkill
- tighten-textSkill
- assess-genai-with-librarySkill
- assessing-genai-securitySkill
- check-styleSkill
- folding-into-the-modelSkill
- humanizerSkill
- keel-designSkill
- tighten-textSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Follow coverage.md. It carries the table of kinds and where each one lands on the chain, the four dispositions, and how to say out_of_scope …
- Distinctive: line 20
Do not skip this because the subject feels new.
- Distinctive: line 26
Follow research.md. It is a procedure with a gate at the end: what you have to find, where to look, and when you are allowed to start …
- Very specific: line 40
Call get_style_guide(entity_type=..., field_name=...) for each field as you write it.
- Distinctive: line 42
Write the card against the control or the chain, never against the one threat that prompted it.
- Distinctive: line 54
If two reasonable modellers would disagree, the human decides.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.
Files
Folding into the model
Placement is the work. Getting the words right is what the style guide is for; deciding what kind of thing this is, whether the model already answers it, whether it is one record or two, and where it attaches is what goes wrong.
Two rules hold the skill together. Do not write from memory - a card is a claim about the world, and the evidence for it has to be found, not recalled. Make the judgment calls visible instead of resolving them silently - communicating your reasoning is not enough, communicate your uncertainty.
Most of what arrives is already answered somewhere, so the skill starts by finding out. Steps 1 and 2 are often the whole job.
1. Decide what kind of thing it is, and whether the model already answers it
Follow coverage.md. It carries the table of kinds and where each one lands on the chain, the four dispositions, and how to say out_of_scope honestly.
It ends in one of two places. Covered or out of scope: record the row and stop - the disposition is the output, there is nothing to author. A gap, whole or partial: carry on here.
Do not skip this because the subject feels new. Skipping it is what produces a near-duplicate of something that was already in the catalog under another name.
2. Research the subject
The step that gets skipped, and the one that shows in the finished card. A card written without it reads as plausible, cites nothing that supports it, and no reviewer can tell it from one written by someone who knew.
Follow research.md. It is a procedure with a gate at the end: what you have to find, where to look, and when you are allowed to start writing.
3. Decide whether it is one record or part of one
The pivot, and the place where a record goes wrong in a way no amount of good writing repairs.
The style guide carries the test in its entity block - the bar for the record as a whole, returned by get_style_guide(entity_type="threat") or get_style_guide(entity_type="mitigation"). A field-scoped call does not return it, so ask for the entity. Read it before you name anything: it also says which decisions come first, and taking them in the wrong order means rewriting the record.
In short. Threats: two chains are two threats if and only if ruling one out does not rule out the other, or closing one does not close the other. Mitigations: two controls are two cards if and only if an organisation could adopt one without the other.
Run the test out loud. Name the reachability and the gating control you have in mind, then say which existing record they match. A subject that fails the test is a weakness on an existing threat, or a criterion on an existing card - which is also the smaller and better change.
4. Write to the bar
Call get_style_guide(entity_type=..., field_name=...) for each field as you write it. One field at a time is far cheaper than the whole entity, and the bar it returns is what the entry will be reviewed against.
Write the card against the control or the chain, never against the one threat that prompted it. A card that carries the shape of its first occasion is not reusable for the second.
5. Say what was debatable
For each of these, name the fork, state your lean, and say how confident you are:
- New record or a change to an existing one. Usually the biggest call, and step 3 is where you made it.
- Which record it attaches to, when two are plausible.
- The fields that carry modelling judgment:
component,surface,harm,nature, and a link'sstrength. - Whether it is in scope at all, if step 1 did not settle it.
- What the research did not find. Absence of evidence is a finding, and it changes how much weight the card can carry.
If two reasonable modellers would disagree, the human decides.
6. Ask before changing meaning
Additive and reversible - a new weakness, a link with a clear rationale - lean toward doing it, and still show the reasoning.
Changing what an existing entry means is not yours to decide: retitling a threat, widening its applicability, re-grading a link's strength, splitting or merging records, deleting anything. Present the change, the why and the trade-off, and get an explicit yes before writing. Never bury one inside an additive change as a "recommendation".
7. Read what the write returns, then get it judged
Every write answers with what it broke. A problems key means this write left the catalog inconsistent - a link pointing at nothing, a gating grade on a control that does not block, a coverage row still claiming something you deleted. Fix it in the same turn. No problems key means there is nothing structural to fix.
That is the deterministic half only. Run the check-style skill on what you just wrote before you report it as done. It is the half that reads the card against its bar, and it is not optional because you feel good about the card.
Where this goes wrong
- Treating a subject as new without checking. Most incoming material is already answered, often on a part of the chain you were not looking at.
- Writing from memory, then adding a plausible-looking reference afterwards. The reference decides the card, not the other way round.
- Presenting a debatable placement as settled. State the fork and your confidence.
- Skipping the splitting test and producing an umbrella - one record covering several chains closed by different controls.
- Naming a card after the threat it was written for, which makes it unusable for the next one.
- Using
out_of_scopefor anything unfamiliar, which quietly turns Keel's boundary into a list of things nobody looked at. - Quietly widening an existing entry's scope as a buried "recommendation". Pull it out and ask.
- Reporting the entry as done on the strength of a clean
keel validate. That gate cannot read a sentence.
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
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.

