context-engineering

Name: context-engineering
Author: tranhieutt

69stars41forks

How to install context-engineering?

tranhieutt/software_development_department/context-engineering

$npx -y skills add tranhieutt/software_development_department --skill context-engineering

Installs into the current project.

Summary

Strictly enforce context engineering principles to avoid context stuffing, optimize memory architecture, and manage the Research-Plan-Reset-Implement cycle.

Files · 1

View on GitHub

SKILL.md

1	# 1. Overview
2	Context engineering bridges the gap between static training data and dynamic reality. Context Stuffing (jamming volume without intent) degrades reasoning, increases noise, and leads to hallucinations. Context Engineering treats AI attention as a scarce resource and allocates it deliberately through structure, bounded contexts, and intelligent memory retrieval (RAG / MCP Supermemory). Without this skill, the AI suffers from "Context Hoarding Disorder," leading to goal drift, high latency, and poor execution quality.
3
4	# 2. When to Use
5	Activate this skill immediately upon detecting the following signs:
6	- Sign 1: User pastes a massive block of uncurated documents entirely into the context window (e.g., full PRDs, full codebases, thousands of lines of logs).
7	- Sign 2: The AI's outputs start to become vague, hedged, or inconsistent despite having "all the context", or when the context window is clearly overflowing.
8	- Sign 3: The user wants to start a multi-step complex workflow spanning many files and iterations.
9	- Implicit Command: User types `/context` or `/memory`.
10
11	# 3. Strict Process
12	ULTIMATUM: You are an Agent. You DO NOT have the right to ignore, truncate, or alter the order of these steps, even if you think "the model has a 1 million token context limit anyway."
13
14	1. [Step 1 - Intent & Boundary Falsification]: Identify exactly what decision the provided context supports. Apply the falsification test: "If I exclude [context element X], what specific failure will occur in [decision Y]?" If there is no concrete failure, the context must be rejected or removed from the active window.
15	2. [Step 2 - Persist vs. Retrieve Classification]: Separate the information. Core constraints and glossary definitions remain in active context. Episodic, project-specific, or historical data must be offloaded and retrieved only when queried. Use `mcp_supermemory_recall` for historical lookups instead of keeping them in the prompt.
16	3. [Step 3 - The R-P-R-I Cycle Execution]:
17	- Research: Gather necessary information.
18	- Plan: Synthesize findings into a high-density `PLAN.md` or `SPEC.md`.
19	- Reset: Save crucial lessons to memory using `mcp_supermemory_memory` and explicitly ask the user to clear the context window (start a new chat) or summarize everything to drop the past context rot.
20	- Implement: Execute purely based on the dense plan.
21	4. [Step 4 - Storage & Consolidation]: Upon finishing a milestone, write the generalized knowledge or operational principles into `mcp_supermemory_memory`.
22
23	# 4. Anti-Rationalizations
24	These are lazy thoughts that Agents (like you) commonly fall prey to. If an idea in your head matches the Left Column, you MUST immediately obey the Right Column:
25
26	\| Agent's Lazy Rationalization \| Refutation & Mandatory Action \|
27	\| :--- \| :--- \|
28	\| "I have a massive context window; I can just read all 50 files the user provided without complaining." \| Volume != Quality. Context is an attention economy. You MUST explicitly tell the user that "Context Stuffing" dilutes attention. Limit the working context to only the files strictly needed for the immediate decision. \|
29	\| "I'll just try again and rewrite the code if it hallucinates." \| Retries mask bad information architecture. Do not normalize retries. Stop and fix the context structure, break down the problem, or refine the retrieved memory BEFORE writing another line of code. \|
30	\| "I'll keep all the research logs and failed attempts in the chat history so I remember what I tried." \| "Context Rot" kills reasoning. Once a plan is synthesized, the previous dead ends become pure noise. You MUST instruct the user to flush the context or start fresh with a clean `SPEC.md`. \|
31
32	# 5. Verification Gates
33	You are NOT allowed to end your turn and respond "Context established" or proceed to implementation until the following checks pass:
34	- [ ] You have successfully filtered out unnecessary broad context and kept only the decision-critical context.
35	- [ ] You have explicitly called `mcp_supermemory_memory` to commit important established facts, or `mcp_supermemory_recall` to fetch past facts instead of asking the user to paste them.
36	- [ ] You have synthesized the current sprawling context into a dense artifact (e.g. `PLAN.md` or `SPEC.md`) and proposed a Context Reset.
37
38	# 6. Red Flags
39	Immediately STOP and request User intervention if:
40	- The user demands that you

People who liked this also viewed

6 picks

Type

context-engineering

69stars41forks

View on GitHub

How to install context-engineering?

tranhieutt/software_development_department/context-engineering

$npx -y skills add tranhieutt/software_development_department --skill context-engineering

Installs into the current project.

Summary

Strictly enforce context engineering principles to avoid context stuffing, optimize memory architecture, and manage the Research-Plan-Reset-Implement cycle.

Files · 1

View on GitHub

SKILL.md

1	# 1. Overview
2	Context engineering bridges the gap between static training data and dynamic reality. Context Stuffing (jamming volume without intent) degrades reasoning, increases noise, and leads to hallucinations. Context Engineering treats AI attention as a scarce resource and allocates it deliberately through structure, bounded contexts, and intelligent memory retrieval (RAG / MCP Supermemory). Without this skill, the AI suffers from "Context Hoarding Disorder," leading to goal drift, high latency, and poor execution quality.
3
4	# 2. When to Use
5	Activate this skill immediately upon detecting the following signs:
6	- Sign 1: User pastes a massive block of uncurated documents entirely into the context window (e.g., full PRDs, full codebases, thousands of lines of logs).
7	- Sign 2: The AI's outputs start to become vague, hedged, or inconsistent despite having "all the context", or when the context window is clearly overflowing.
8	- Sign 3: The user wants to start a multi-step complex workflow spanning many files and iterations.
9	- Implicit Command: User types `/context` or `/memory`.
10
11	# 3. Strict Process
12	ULTIMATUM: You are an Agent. You DO NOT have the right to ignore, truncate, or alter the order of these steps, even if you think "the model has a 1 million token context limit anyway."
13
14	1. [Step 1 - Intent & Boundary Falsification]: Identify exactly what decision the provided context supports. Apply the falsification test: "If I exclude [context element X], what specific failure will occur in [decision Y]?" If there is no concrete failure, the context must be rejected or removed from the active window.
15	2. [Step 2 - Persist vs. Retrieve Classification]: Separate the information. Core constraints and glossary definitions remain in active context. Episodic, project-specific, or historical data must be offloaded and retrieved only when queried. Use `mcp_supermemory_recall` for historical lookups instead of keeping them in the prompt.
16	3. [Step 3 - The R-P-R-I Cycle Execution]:
17	- Research: Gather necessary information.
18	- Plan: Synthesize findings into a high-density `PLAN.md` or `SPEC.md`.
19	- Reset: Save crucial lessons to memory using `mcp_supermemory_memory` and explicitly ask the user to clear the context window (start a new chat) or summarize everything to drop the past context rot.
20	- Implement: Execute purely based on the dense plan.
21	4. [Step 4 - Storage & Consolidation]: Upon finishing a milestone, write the generalized knowledge or operational principles into `mcp_supermemory_memory`.
22
23	# 4. Anti-Rationalizations
24	These are lazy thoughts that Agents (like you) commonly fall prey to. If an idea in your head matches the Left Column, you MUST immediately obey the Right Column:
25
26	\| Agent's Lazy Rationalization \| Refutation & Mandatory Action \|
27	\| :--- \| :--- \|
28	\| "I have a massive context window; I can just read all 50 files the user provided without complaining." \| Volume != Quality. Context is an attention economy. You MUST explicitly tell the user that "Context Stuffing" dilutes attention. Limit the working context to only the files strictly needed for the immediate decision. \|
29	\| "I'll just try again and rewrite the code if it hallucinates." \| Retries mask bad information architecture. Do not normalize retries. Stop and fix the context structure, break down the problem, or refine the retrieved memory BEFORE writing another line of code. \|
30	\| "I'll keep all the research logs and failed attempts in the chat history so I remember what I tried." \| "Context Rot" kills reasoning. Once a plan is synthesized, the previous dead ends become pure noise. You MUST instruct the user to flush the context or start fresh with a clean `SPEC.md`. \|
31
32	# 5. Verification Gates
33	You are NOT allowed to end your turn and respond "Context established" or proceed to implementation until the following checks pass:
34	- [ ] You have successfully filtered out unnecessary broad context and kept only the decision-critical context.
35	- [ ] You have explicitly called `mcp_supermemory_memory` to commit important established facts, or `mcp_supermemory_recall` to fetch past facts instead of asking the user to paste them.
36	- [ ] You have synthesized the current sprawling context into a dense artifact (e.g. `PLAN.md` or `SPEC.md`) and proposed a Context Reset.
37
38	# 6. Red Flags
39	Immediately STOP and request User intervention if:
40	- The user demands that you

People who liked this also viewed

6 picks

Type