Qi layer skill

Use when writing or maintaining AGENTS.md, .context/CONTEXT.md, or CLAUDE.md mirrors: keep intent docs minimal and load-bearing.

by haowjy·Apache-2.0 license·★ 506 Stars on the repo·GitHub ↗

Use now

haowjy/main1 file shown

SKILL.md5.3 KB

Browse every file on GitHub↗

SKILL.md

Show the full text118 lines

qi-layer/SKILL.md118 lines · 5.3 KB

PreviewCode

Outline

RawView on GitHub

qi-layer

/qi-maintenance owns when colocated knowledge must move with source changes. This skill owns how to write and structure that knowledge.

Load /knowledge-layers for where each layer lives and what it holds. Load /llm-writing if it isn't already loaded.

This skill is about how to write and maintain the directory-local pair: AGENTS.md and .context/CONTEXT.md. The pair governs any tree agents work in — code, the KB, docs, work directories. AGENTS.md loads into an agent's bounded context as standing instructions: write it like a prompt, minimal, every line load-bearing.

The Four Principles

Fractal Compression: leaf AGENTS.md summarizes its directory's content; parent AGENTS.md summarizes its children. Each level is a compression of the level below.
Hierarchical Summarization: root provides broad architectural frame. Leaves provide local working knowledge. Agents accumulate understanding as they descend.
LCA Deduplication: shared knowledge appears once at the shallowest node covering all relevant paths. Never duplicate between siblings.
Progressive Disclosure: give just enough to work correctly at this level. Link to .context/CONTEXT.md for depth.

Writing AGENTS.md

Agents read AGENTS.md before opening anything else in the tree — write for that moment. Ask: what must someone understand before working here? That's what AGENTS.md captures.

Keep AGENTS.md as short as the directory allows, rarely past 200 lines. Include only what has substance:

Purpose: what this area IS and what it ISN'T (1–3 sentences)
Mental model: how to think about this area, key abstractions
Key rules: constraints, what breaks if you get it wrong
Anti-patterns: what NOT to do here
Downlinks: to .context/ for depth, to related areas

An agent that only reads AGENTS.md should be able to work correctly here. An agent that also reads .context/ should be able to change things safely.

Writing .context/CONTEXT.md

Reference depth, co-located with the code it describes. Where an agent goes when it needs contracts, architecture, or rationale in detail.

Sections (use only those with substance):

Contracts: interfaces, invariants, what breaks if violated
Architecture: component relationships, data flow, dependency direction
Rationale: why X over Y, rejected alternatives
Patterns: how to work here, concrete pitfalls

The .context/ directory is extensible: additional files alongside CONTEXT.md for specialized concerns.

What Does NOT Belong in AGENTS.md

Apply the every-session test: root AGENTS.md loads on every session. If knowledge is only relevant when working in a specific domain, it belongs in that domain's AGENTS.md or .context/, not root.

Apply the think-vs-lookup test: text whose removal would cause a wrong decision belongs in AGENTS.md. Text an agent would merely have to look up belongs in .context/. Text that changes no behavior gets deleted — agents already know how to code and follow common conventions.

Specific failure modes:

Session bleed: LLM working notes that calcified into the instruction file. Tells: status-update language ("deleted", "shipped", "deferred"), implementation terms packed without framing, history narration.
Reference material posing as intent: tables, command blocks, full scheme vocabularies, implementation specifics.
Redundant guards: prose warnings for invariants already enforced by tests or types. The code is the real guard; prose rots faster.
Duplicated knowledge: restating what lives in a skill, a domain .context/, or a KB page. Point, don't duplicate.
Domain-specific detail at root: URI scheme tables, gateway pricing internals, auth implementation details. These belong in their domain's AGENTS.md, not root.

Structural Rules

Relative paths for all links
AGENTS.md and .context/ at the same directory level (siblings)
Link to files, not headings (headings change more often)
Lateral links between .context/ directories with contracts between them
LCA deduplication: if two siblings share context, put it in the parent

CLAUDE.md Mirrors

Claude harnesses read CLAUDE.md, not AGENTS.md. Give every AGENTS.md a sibling CLAUDE.md whose first line is @AGENTS.md — normally the whole file. Run meridian qi claude-md-fix <target-root> on the containing tree after creating or moving AGENTS.md files: it creates missing mirrors, skips exact ones, and reports anything else as a conflict.

Never write shared instructions into CLAUDE.md. Claude-only knowledge is rare; when it exists, put it below the @AGENTS.md import and expect claude-md-fix to keep flagging the file, so the divergence stays visible.

Loading differs by level. At the root, each harness auto-loads its own file every session: Claude reads CLAUDE.md, others read AGENTS.md. In subdirectories, Claude auto-injects CLAUDE.md when it touches files there; other agents see nested AGENTS.md only by reading it on entry. Don't lean on Claude's auto-injection: a nested AGENTS.md carries the local additions an agent needs on entry, with everything else inherited from ancestors.

1	---
2	name: qi-layer
3	description: 'Use when writing or maintaining AGENTS.md, .context/CONTEXT.md, or CLAUDE.md mirrors: keep intent docs minimal and load-bearing.'
4	---
5
6	# qi-layer
7
8	`/qi-maintenance` owns when colocated knowledge must move with source changes.
9	This skill owns how to write and structure that knowledge.
10
11	Load `/knowledge-layers` for where each layer lives and what it holds.
12	Load `/llm-writing` if it isn't already loaded.
13
14	This skill is about how to write and maintain the directory-local pair:
15	AGENTS.md and `.context/CONTEXT.md`. The pair governs any tree agents work
16	in — code, the KB, docs, work directories. AGENTS.md loads into an agent's
17	bounded context as standing instructions: write it like a prompt, minimal,
18	every line load-bearing.
19
20	## The Four Principles
21
22	1. Fractal Compression: leaf AGENTS.md summarizes its directory's
23	content; parent AGENTS.md summarizes its children. Each level is a
24	compression of the level below.
25	2. Hierarchical Summarization: root provides broad architectural
26	frame. Leaves provide local working knowledge. Agents accumulate
27	understanding as they descend.
28	3. LCA Deduplication: shared knowledge appears once at the shallowest
29	node covering all relevant paths. Never duplicate between siblings.
30	4. Progressive Disclosure: give just enough to work correctly at this
31	level. Link to `.context/CONTEXT.md` for depth.
32
33	## Writing AGENTS.md
34
35	Agents read AGENTS.md before opening anything else in the tree — write for
36	that moment. Ask: what must someone understand before working here?
37	That's what AGENTS.md captures.
38
39	Keep AGENTS.md as short as the directory allows, rarely past 200 lines.
40	Include only what has substance:
41
42	- Purpose: what this area IS and what it ISN'T (1–3 sentences)
43	- Mental model: how to think about this area, key abstractions
44	- Key rules: constraints, what breaks if you get it wrong
45	- Anti-patterns: what NOT to do here
46	- Downlinks: to `.context/` for depth, to related areas
47
48	An agent that only reads AGENTS.md should be able to work correctly here.
49	An agent that also reads .context/ should be able to change things safely.
50
51	## Writing .context/CONTEXT.md
52
53	Reference depth, co-located with the code it describes. Where an agent
54	goes when it needs contracts, architecture, or rationale in detail.
55
56	Sections (use only those with substance):
57
58	- Contracts: interfaces, invariants, what breaks if violated
59	- Architecture: component relationships, data flow, dependency direction
60	- Rationale: why X over Y, rejected alternatives
61	- Patterns: how to work here, concrete pitfalls
62
63	The `.context/` directory is extensible: additional files alongside
64	CONTEXT.md for specialized concerns.
65
66	## What Does NOT Belong in AGENTS.md
67
68	Apply the every-session test: root AGENTS.md loads on every session.
69	If knowledge is only relevant when working in a specific domain, it belongs
70	in that domain's AGENTS.md or .context/, not root.
71
72	Apply the think-vs-lookup test: text whose removal would cause a
73	wrong decision belongs in AGENTS.md. Text an agent would merely have to
74	look up belongs in .context/. Text that changes no behavior gets
75	deleted — agents already know how to code and follow common conventions.
76
77	Specific failure modes:
78
79	- Session bleed: LLM working notes that calcified into the instruction
80	file. Tells: status-update language ("deleted", "shipped", "deferred"),
81	implementation terms packed without framing, history narration.
82	- Reference material posing as intent: tables, command blocks, full
83	scheme vocabularies, implementation specifics.
84	- Redundant guards: prose warnings for invariants already enforced by
85	tests or types. The code is the real guard; prose rots faster.
86	- Duplicated knowledge: restating what lives in a skill, a domain
87	.context/, or a KB page. Point, don't duplicate.
88	- Domain-specific detail at root: URI scheme tables, gateway pricing
89	internals, auth implementation details. These belong in their domain's
90	AGENTS.md, not root.
91
92	## Structural Rules
93
94	- Relative paths for all links
95	- AGENTS.md and .context/ at the same directory level (siblings)
96	- Link to files, not headings (headings change more often)
97	- Lateral links between `.context/` directories with contracts between them
98	- LCA deduplication: if two siblings share context, put it in the parent
99
100	## CLAUDE.md Mirrors
101
102	Claude harnesses read CLAUDE.md, not AGENTS.md. Give every AGENTS.md a
103	sibling CLAUDE.md whose first line is `@AGENTS.md` — normally the whole
104	file. Run `meridian qi claude-md-fix <target-root>` on the containing tree
105	after creating or moving AGENTS.md files: it creates missing mirrors, skips
106	exact ones, and reports anything else as a conflict.
107
108	Never write shared instructions into CLAUDE.md. Claude-only knowledge is
109	rare; when it exists, put it below the `@AGENTS.md` import and expect
110	`claude-md-fix` to keep flagging the file, so the divergence stays visible.
111
112	Loading differs by level. At the root, each harness auto-loads its own
113	file every session: Claude reads CLAUDE.md, others read AGENTS.md. In
114	subdirectories, Claude auto-injects CLAUDE.md when it touches files there;
115	other agents see nested AGENTS.md only by reading it on entry. Don't lean
116	on Claude's auto-injection: a nested AGENTS.md carries the local additions
117	an agent needs on entry, with everything else inherited from ancestors.
118