Skip to content

Commit a99f8fd

Browse files
enunoenuno
committed
docs: add MEMORY.md with curated Dokploy patterns and refresh CLAUDE.md with Cloudflare integration
- Created MEMORY.md: Architectural decisions, implementation patterns, debugging playbooks, DevOps rules - Updated CLAUDE.md v2.1.0: Added template patterns, Cloudflare checklist, creation workflow - Added claude-mem-mastery skill files for maintaining project memory - Extracted lessons from ai-context template (PR #6): single-service pattern, progressive skill loading, clarification workflow
1 parent d807943 commit a99f8fd

7 files changed

Lines changed: 1709 additions & 2 deletions

File tree

‎.claude/skills/claude-mem-mastery/SKILL.md‎

Lines changed: 173 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,173 @@
1+
---
2+
name: claude-mem-coded-assistant
3+
description: >
4+
Entry-point skill for using claude-mem to keep CLAUDE.md and MEMORY.md
5+
in sync so Claude learns from past work and avoids repeating mistakes.
6+
version: 1.1.0
7+
---
8+
9+
# Claude‑Mem Coding Skill
10+
11+
## What This Skill Does
12+
13+
This skill teaches Claude how to:
14+
15+
- Mine **claude-mem** (via MCP) for high‑signal past work.
16+
- Maintain a concise, high‑impact **CLAUDE.md** (~1,500 tokens).
17+
- Maintain a curated **MEMORY.md** of lessons learned and directions, so future work is faster and less error‑prone.
18+
19+
It is an **entry point**, not a full manual. Detailed workflows and examples live in separate reference files that Claude can open on demand.
20+
21+
---
22+
23+
## When to Use This Skill
24+
25+
Claude should activate this skill when:
26+
27+
- A feature, refactor, or significant bugfix is completed.
28+
- An infra/deployment change introduces new operational lessons.
29+
- Starting work on an area with substantial history in claude-mem.
30+
- Performing a daily “memory maintenance” pass on an active repo.
31+
32+
---
33+
34+
## Inputs and Outputs
35+
36+
### Inputs
37+
38+
Claude relies on:
39+
40+
- **Files** (in repo root):
41+
- `CLAUDE.md` – main project instructions.
42+
- `MEMORY.md` – curated lessons and directions.
43+
- **claude-mem MCP tools** (already installed & connected):
44+
- `search` – index‑level observation search.
45+
- `timeline` – temporal context around observations.
46+
- `get_observations` – full structured details.
47+
48+
### Outputs
49+
50+
This skill produces:
51+
52+
- **Patch‑style edits** to:
53+
- `MEMORY.md` – new or updated lessons, patterns, and playbooks.
54+
- `CLAUDE.md` – refreshed rules while staying under ~1,500 tokens.
55+
- No raw claude-mem transcripts are copied; only compressed, actionable guidance.
56+
57+
---
58+
59+
## How Claude Should Behave
60+
61+
### 1. Mine claude-mem → Update MEMORY.md
62+
63+
High‑level behavior (details in `claude-mem-usage.md`):
64+
65+
- Use **progressive disclosure** against claude-mem:
66+
1. `search` for recent `decision`, `bugfix`, `refactor`, `discovery`, `change` observations.
67+
2. `timeline` around promising IDs to see context.
68+
3. `get_observations` for a small set of high‑value IDs.
69+
- From those, update `MEMORY.md` with:
70+
- Architectural decisions and their impact.
71+
- Implementation patterns and anti‑patterns.
72+
- Debugging playbooks and DevOps lessons.
73+
74+
**Constraints**
75+
76+
- Prefer short bullets over long prose.
77+
- Record *why* decisions were made and how to act next time.
78+
- Never store secrets or credentials in `MEMORY.md`.
79+
80+
For a full template and examples, Claude should open:
81+
82+
- `memory-structure-reference.md`
83+
- `claude-mem-usage.md`
84+
85+
---
86+
87+
### 2. Distill MEMORY.md → Refresh CLAUDE.md (≈1,500 tokens)
88+
89+
High‑level behavior:
90+
91+
- Read the existing `CLAUDE.md` and approximate its size; keep the body around **1–1.5k tokens** for optimal behavior.
92+
- Pull only **current, high‑impact** content from `MEMORY.md`:
93+
- Still‑valid architectural directions.
94+
- Frequently reused patterns and gotchas.
95+
- Operational guardrails that materially affect daily work.
96+
- Rewrite historical notes as **timeless rules**, e.g.:
97+
- “When adding retries to DB writes, always use the shared retry helper instead of manual loops.”
98+
99+
- Use links instead of inlining:
100+
- `.clauderules/code-style.md` for style.
101+
- `.clauderules/testing.md` for testing.
102+
- `MEMORY.md` sections for deeper background.
103+
104+
**Token Discipline**
105+
106+
- If CLAUDE.md is too long:
107+
- Merge overlapping bullets.
108+
- Drop generic advice that doesn’t change behavior.
109+
- Replace detailed explanations with references to supporting docs.
110+
111+
**Diff‑First**
112+
113+
- Propose **minimal patches**, not full rewrites:
114+
- Update only sections that need change (e.g., “Architectural Directions”, “Patterns & Gotchas”).
115+
- Preserve stable layout and headings.
116+
- Always leave final acceptance to human review in Git/CI.
117+
118+
For concrete layouts and example diffs, Claude should open:
119+
120+
- `claude-md-layout-reference.md`
121+
- `example-diffs.md`
122+
123+
---
124+
125+
## Safety and Priority Rules
126+
127+
Claude must:
128+
129+
- **Always**:
130+
- Query claude-mem before re‑solving problems already encountered in this project.
131+
- Update `MEMORY.md` after meaningful work with concise, actionable lessons.
132+
- Keep `CLAUDE.md` focused on rules that change how work is done, not on general LLM tips.
133+
134+
- **Never**:
135+
- Overwrite `CLAUDE.md` or `MEMORY.md` entirely; always propose small diffs.
136+
- Paste raw claude-mem observations verbatim into either file.
137+
- Store secrets, API keys, or sensitive infra details in these files.
138+
139+
- **Conflict resolution priority**:
140+
1. Explicit instructions in `CLAUDE.md`.
141+
2. Latest curated guidance in `MEMORY.md`.
142+
3. Raw claude-mem observations and session summaries.
143+
4. Ad‑hoc reasoning in the current session.
144+
145+
---
146+
147+
## Quick “How to Call Me”
148+
149+
Users can invoke this skill with prompts like:
150+
151+
> “Use the claude-mem coding skill to:
152+
> 1) mine claude-mem for recent work,
153+
> 2) update MEMORY.md with lessons, and
154+
> 3) refresh CLAUDE.md under the ~1,500‑token budget.”
155+
156+
Claude should then:
157+
158+
1. Run the claude-mem `search → timeline → get_observations` flow.
159+
2. Draft a patch for `MEMORY.md` with new lessons.
160+
3. Draft a patch for `CLAUDE.md` derived from `MEMORY.md`.
161+
4. Present both patches clearly for human review and commit.
162+
163+
---
164+
165+
## External References
166+
167+
To keep this SKILL.md lean and within best‑practice size, Claude should open these files when more detail is needed:
168+
169+
- `claude-mem-usage.md` – detailed claude-mem MCP workflows, filters, and example queries.
170+
- `memory-structure-reference.md` – full MEMORY.md templates and longer examples.
171+
- `claude-md-layout-reference.md` – canonical CLAUDE.md section layouts and size guidance.
172+
- `example-diffs.md` – sample before/after patches for CLAUDE.md and MEMORY.md.
173+

0 commit comments

Comments
 (0)