notes: renumber Engineering Notes companion 06 → 05 (prefix cache) (#9)

* notes: renumber context-tax companion to Engineering Notes 05

Article 05 is now the prefix-cache note (was planned as 06).
Rename the companion note, drop the upcoming marker, and retarget
skill/checklist references.

* Align context-tax checklist with tightened note 05

Drop mid-session mode-consistency items; the note now drills one
point: prefix cache position and precision.

* notes(05): retarget example to prefix placement, not request-tail thrift

Align the note, checklist, and context-tax-audit skill with the article:
volatile time stays out of the system prompt; drill prefix protection.

* notes(05): retell via how the agent learns current time

Align companion note, checklist and skill with the time→SP→prefix-miss→
MCode placement arc; drop the old 'one small field' abstract lead.

* notes(05): sync title and blurb with Feishu time-placement arc

Match the polished article: how the agent learns current time; history
write-once for <agent-context>; README title aligned.

* chore(ronny): bump stack version to 0.2.5

Checklist/skill updates for notes 05 need a marketplace-visible version bump.

Author: hetaoBackend

README.md

@@ -182,7 +182,7 @@ change it. To suggest a change here, open an issue; see
 | 02 | [Too many tools will eat your context: from tool search to PTC](notes/02-tool-context.md) | `tool-budget-audit` |
 | 03 | [Keeping long sessions sharp: compaction vs. memory](notes/03-compaction-memory.md) | `long-session-check`, `handoff-doc` |
 | 04 | [How long should one bash command wait: from foreground blocking to background tasks](notes/04-bash-background.md) | `long-command-check` |
-| 06 | [The hidden tax on every turn: what an agent puts into context](notes/06-context-tax.md) *(upcoming)* | `context-tax-audit` |
+| 05 | [One small field can break the whole prefix cache](notes/05-prefix-cache.md) | `context-tax-audit` |
 
 ## Stacks
 

README_ZH.md

@@ -142,7 +142,7 @@ AI 写代码越来越快，真正拉开差距的，是围绕 agent 的那一圈
 | 02 | [工具一多，context 就爆：从 tool search 到 PTC](notes/02-tool-context.md) | `tool-budget-audit` |
 | 03 | [长会话怎么不变傻：compaction 和 memory 的取舍](notes/03-compaction-memory.md) | `long-session-check`、`handoff-doc` |
 | 04 | [一条 bash 命令该等多久：从前台阻塞到后台任务](notes/04-bash-background.md) | `long-command-check` |
-| 06 | [每一轮都在交的隐形税：agent 往上下文里塞了什么](notes/06-context-tax.md)（即将发布） | `context-tax-audit` |
+| 05 | [一个小字段，拖垮整段前缀缓存](notes/05-prefix-cache.md) | `context-tax-audit` |
 
 ## Stacks
 

notes/05-prefix-cache.md

@@ -0,0 +1,21 @@
+# 05 · How does the agent know the current time?
+
+**In one line:** the model has no clock. Stuffing `date` into the system
+prompt is the most direct fix — and it breaks the prefix cache every turn.
+Put session-stable `# Environment` up front; put per-turn time in
+`<agent-context>` that is written into history and never rewritten; keep
+precision to the minute. For each injected part, ask what breaks without it,
+whether it needs every turn, how precise it must be, and whether its position
+breaks the prompt cache.
+
+**Companion skill:** `context-tax-audit` - inventory one real request, then
+check necessity, frequency, precision, position and cache impact.
+
+Plain-text checklist: [`checklists/context-tax.md`](../stacks/ronny/checklists/context-tax.md)
+
+**Install:**
+
+```bash
+mkdir -p ~/.agents/skills
+cp -R ~/mcode-stack/stacks/ronny/skills/context-tax-audit ~/.agents/skills/
+```

notes/06-context-tax.md

@@ -10,6 +10,6 @@ repository holds something you can install and run for it.
 | 02 | [Too many tools will eat your context: from tool search to PTC](02-tool-context.md) | `tool-budget-audit` |
 | 03 | [Keeping long sessions sharp: compaction vs. memory](03-compaction-memory.md) | `long-session-check`, `handoff-doc` |
 | 04 | [How long should one bash command wait: from foreground blocking to background tasks](04-bash-background.md) | `long-command-check` |
-| 06 | [The hidden tax on every turn: what an agent puts into context](06-context-tax.md) *(upcoming)* | `context-tax-audit` |
+| 05 | [How does the agent know the current time?](05-prefix-cache.md) | `context-tax-audit` |
 
 All companion skills live in [`stacks/ronny/skills/`](../stacks/ronny/skills/).

notes/README.md

@@ -1,7 +1,7 @@
 {
   "name": "ronny",
   "displayName": "Ronny's MCode Stack",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "description": "Daily workflows and self-check skills from the MiniMax Code R&D lead: independent review, subagent split checks, tool and context budget audits, long sessions, long commands, harness experiments, trajectory review, plus everyday engineering practice for code review, review comments, small PRs, change descriptions and emergency changes.",
   "author": {
     "name": "Ronny (MiniMax Code team)"

stacks/ronny/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "ronny",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "description": "Daily workflows and self-check skills from the MiniMax Code R&D lead: independent review, subagent split checks, tool and context budget audits, long sessions, long commands, harness experiments, trajectory review, plus everyday engineering practice for code review, review comments, small PRs, change descriptions and emergency changes.",
   "author": {
     "name": "Ronny (MiniMax Code team)"
@@ -14,7 +14,10 @@
     "longDescription": "Daily workflows and self-check skills from the MiniMax Code R&D lead: independent review, subagent split checks, tool and context budget audits, long sessions, long commands, harness experiments, trajectory review, plus everyday engineering practice for code review, review comments, small PRs, change descriptions and emergency changes.",
     "developerName": "Ronny (MiniMax Code team)",
     "category": "Developer Tools",
-    "capabilities": ["Read", "Write"],
+    "capabilities": [
+      "Read",
+      "Write"
+    ],
     "websiteURL": "https://github.com/MiniMax-AI/mcode-stack/tree/main/stacks/ronny",
     "defaultPrompt": [
       "Review this diff from a clean context against the requirement",

stacks/ronny/.codex-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
   "name": "ronny",
-  "version": "0.2.4",
+  "version": "0.2.5",
   "displayName": "Ronny's MCode Stack",
   "description": "Daily workflows and self-check skills from the MiniMax Code R&D lead: independent review, subagent split checks, tool and context budget audits, long sessions, long commands, harness experiments, trajectory review, plus everyday engineering practice for code review, review comments, small PRs, change descriptions and emergency changes.",
   "author": "Ronny (MiniMax Code team)",

stacks/ronny/.minimax-plugin/plugin.json

@@ -42,12 +42,12 @@ shown by `mcode plugin marketplace list`, or copy single skills into
 | --- | --- | --- |
 | [`independent-review`](skills/independent-review/SKILL.md) | Review a diff from a clean context: only diff, requirement and test results; structured findings verified before fixing | 01 |
 | [`subagent-split-check`](skills/subagent-split-check/SKILL.md) | Decide whether work goes to a subagent or stays in the main loop; merge existing subagents back step by step | 01 |
-| [`harness-experiment`](skills/harness-experiment/SKILL.md) | Run a harness change as an experiment: small eval, one change, repeated runs, flipped tasks | 01, 02, 06 |
+| [`harness-experiment`](skills/harness-experiment/SKILL.md) | Run a harness change as an experiment: small eval, one change, repeated runs, flipped tasks | 01, 02, 05 |
 | [`tool-budget-audit`](skills/tool-budget-audit/SKILL.md) | Measure what tools cost in context and pick between trimming, tool search and programmatic tool calling | 02 |
 | [`long-session-check`](skills/long-session-check/SKILL.md) | Drop, summarize or write to file; when to compact vs. start fresh; review compaction and memory design | 03 |
 | [`handoff-doc`](skills/handoff-doc/SKILL.md) | Write a hand-off doc (goal, decisions, ruled out, next step) so a new session can pick up | 01, 03 |
 | [`long-command-check`](skills/long-command-check/SKILL.md) | Ten-point audit of how an agent runs long shell commands and background tasks | 04 |
-| [`context-tax-audit`](skills/context-tax-audit/SKILL.md) | Inventory what is injected into every request and how it affects the prompt cache | 06 |
+| [`context-tax-audit`](skills/context-tax-audit/SKILL.md) | Inventory what is injected into every request and how it affects the prompt cache | 05 |
 | [`trajectory-review`](skills/trajectory-review/SKILL.md) | Review how the agent got its result, not only whether it passed | - |
 
 ### Engineering practice

stacks/ronny/README.md

@@ -1,16 +1,14 @@
 # Checklist: what you inject into every request
 
-From MCode Engineering Notes 06 (upcoming). Skill version: `skills/context-tax-audit`.
+From MCode Engineering Notes 05. Skill version: `skills/context-tax-audit`.
 
 | Item | Ask yourself |
 | --- | --- |
-| Inventory | Can you list everything in a request besides the user message, and how many tokens each part takes? |
+| Prefix awareness | Can you point to, in one request, what is the stable prefix and what changes every turn? |
 | Necessity | If you delete each part, where would the model go wrong? If you cannot say, consider deleting it. |
-| Frequency | Does it need to appear in every turn and every request? Could it appear only at the start, at key points, or on change? |
+| Position | Is current time (or anything else that changes) in the system prompt? Move it later — e.g. time in per-turn `<agent-context>`, not `# Environment`. |
+| Frequency | Does a changing field need every turn / every request? Could it appear only at the start, at key points, or on change? |
 | Precision | Does time need seconds? Do numbers need every digit? The more precise, the more it changes every turn. |
-| Position | Is anything that changes placed in the system prompt or tool definitions? Move it later. |
 | Don't edit history | Are reminders already written into history rewritten in later turns? |
-| Request-level reminders | Reminders attached only to the end of a request are paid at full price every time. Is their frequency controlled? |
 | Recovery | Does removed information disappear entirely on compaction, session resume or subagents? Who brings it back? |
-| Mode consistency | Can the prompt or tool set switch mid-session? |
 | Measurement | Before and after a change, do you look at cache hit rate, cost per turn and quality together? Token count alone is not enough. |