matching-references
Skill8 starsChanged 4 months agoMITView on GitHub
Identifies sources in the corpus that match a query topic, situation, or concept. Returns a list with one-line justifications for each candidate, with paths to the relevant distillations.
In this repository15 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- advise-from-corpusSkill
- answer-from-corpusSkill
- audit-attributionSkill
- creating-applicationsSkill
- creating-corpusSkill
- creating-distillationsSkill
- creating-lensesSkill
- creating-tasksSkill
- finding-resourcesSkill
- gridmaker-interviewSkill
- ingesting-imagesSkill
- ingesting-resourcesSkill
- removing-resourcesSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 13
Path convention: Every path in this skill is written from the app's perspective: ./distillations/{task}/{slug}-{task}.md, …
- Distinctive: line 21
No-coverage is honest: if none of the above surfaces a match, say so plainly rather than fabricate from training.
- Very specific: line 25
Read ./slug-table.json, and the concept index the way its schema needs: the topics block on schema 3, the whole file on schema 2.
≈ “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
Matching References
Identifies which sources in the corpus address a given query topic, situation, or concept. The skill returns a list of candidate sources with one-line justifications, naming the distillation(s) that project each source onto the task axes the app ships.
Distillations are the source-grounded product. The reference tier (light + deep) lives at corpus level as the audit-of-record but does not travel with the compiled app, so this skill returns distillation paths rather than reference paths.
Path convention. Every path in this skill is written from the app's perspective: ./distillations/{task}/{slug}-{task}.md, ./concept-index.json, ./slug-table.json. The distillations directory and the runtime JSON indexes sit at the app root.
Retrieval order
- Runtime listener tables in the distillation index for the task at hand (the
… — listenersections of./distillations/{task}/task-index.json, one per phase; the{TASK}-DISTILLATION-INDEX.mdis the operator view of the same tables). In a split index (schema_version2) a listener section lists each trigger with its[D#]pointers and names ashardfile holding the full trigger→response rows: match the trigger in the main file, then load only that shard. Check first when the query carries observable triggers: an utterance, a dynamic, a signal the practitioner noticed in the room ("the team keeps saying X", "we're stuck on Y", "someone keeps doing Z"). Skip this step on task axes whose spec carries no field-2a seed table. - Phase-routing tables in the task-axis index (
./distillations/{task}/task-index.json,rowsfield). Use when the query is at the level of a phase or situation type ("we're in the contributing-factor-analysis phase", "this is a stakeholder-management question") rather than an observable trigger. - Concept axis (
./concept-index.json) for topic- and concept-shaped queries. A schema-3 index opens with a topics block, the subjects the library covers with their synonyms: pick the topics the query touches and fetch their rows by topic ID, asanswer-from-corpussets out under Reading the concept index. On schema 2, each row is a concept with its aliases and the sources that cover it. - Slug-table scan (
./slug-table.json) for named-source / author / title lookups when the user names a specific work or author. Resolve the slug-ID to distillation paths via./distillations/{task}/{slug}-{task}.mdfor each task the app ships. - No-coverage is honest: if none of the above surfaces a match, say so plainly rather than fabricate from training.
Procedure
- Read
./slug-table.json, and the concept index the way its schema needs: the topics block on schema 3, the whole file on schema 2. - Match the query against the concept index first for concept-shaped queries (topic names and synonyms, then the rows under the matching topics; on schema 2, concept names and aliases); against
slug-table.json's slugs for source-shaped queries (author/title strings often appear in the slug itself). - For each matched slug, find the distillation files that project it onto the task axes this app ships (look under
./distillations/). - Return the matches as a list with paths to the corresponding distillation files, in relevance order.
Inputs
- A query: topic, concept, or situation in plain language.
Outputs
- A list of candidate sources. For each:
- Author / topic slug
- Path(s) to the distillation(s) for the task axes the app ships
- One-line justification (why this source matches the query)
When to use
- The user asks "is there a source in this corpus on X?"
- The user describes a situation and wants the assistant to identify which sources apply.
When not to use
- The user has already named the source and the task. Read the distillation directly; do not run a search to find what is already named.
- The user has named a phase of a task. Use the distillation index's phase-routing tables for that task instead. The distillation index is the situation-to-resource router; this skill is the topic-to-resource router.
- The user describes an observable trigger from the room (an utterance, a dynamic, an energy signal: "the team keeps saying X", "we're stuck arguing about Y"). Use the distillation index's runtime listener tables for the task at hand. The listener table is the moment-to-framework router and ships in any application whose task spec carries a field-2a seed table.
Related skills
answer-from-corpus: the default source-grounded retrieval skill.
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.

