context-engine

Name: context-engine
Author: Rune-kit

80stars21forks

How to install context-engine?

rune-kit/rune/context-engine

$curl -o .claude/agents/context-engine.md https://raw.githubusercontent.com/Rune-kit/rune/HEAD/agents/context-engine.md

Installs into the current project.

Summary

Context window management — tracks tool call count, auto-detects when context fills up, triggers state save + compaction. Artifact folding for large outputs (>4000 chars → file + preview).

Files · 1

View on GitHub

agents/context-engine.md

1	# context-engine
2
3	## Purpose
4
5	Context window management for long sessions. Detects when context is approaching limits, triggers smart compaction preserving critical decisions and progress, and coordinates with session-bridge to save state before compaction. Prevents the common failure mode of losing important context mid-workflow.
6
7	### Behavioral Contexts
8
9	Context-engine also manages behavioral mode injection via `contexts/` directory. Three modes are available:
10
11	\| Mode \| File \| When to Use \|
12	\|------\|------\|-------------\|
13	\| `dev` \| `contexts/dev.md` \| Active coding — bias toward action, code-first \|
14	\| `research` \| `contexts/research.md` \| Investigation — read widely, evidence-based \|
15	\| `review` \| `contexts/review.md` \| Code review — systematic, severity-labeled \|
16
17	Mode activation: Orchestrators (cook, team, rescue) can set the active mode by writing to `.rune/active-context.md`. The session-start hook injects the active context file into the session. Mode switches mid-session are supported — the orchestrator updates the file and references the new behavioral rules.
18
19	Default: If no `.rune/active-context.md` exists, no behavioral mode is injected (standard Claude behavior).
20
21	## Triggers
22
23	- Called by `cook` and `team` automatically at context boundaries
24	- Auto-trigger: when tool call count exceeds threshold or context utilization is high
25	- Auto-trigger: before compaction events
26
27	## Calls (outbound)
28
29	# Exception: L3→L3 coordination
30	- `session-bridge` (L3): coordinate state save when context critical
31
32	## Called By (inbound)
33
34	- `cook` (L1): Phase boundaries and when tool count exceeds thresholds
35	- `team` (L1): before parallel workstream dispatch, after merge
36	- `rescue` (L1): between refactoring sessions for state persistence
37	- `context-pack` (L3): when packaging context for sub-agent handoff
38	- `session-bridge` (L3): coordinates with context-engine for compaction timing
39	- `adversary` (L2): (oracle-mode) emit `context.preview` before bundle build to gate token cost
40
41	## Execution
42
43	### Step 1 — Count tool calls
44
45	Count total tool calls made so far in this session. This is the ONLY reliable metric — token usage is not exposed by Claude Code and any estimate will be dangerously inaccurate.
46
47	Do NOT attempt to estimate token percentages. Tool count is a directional proxy, not a precise measurement.
48
49	### Step 2 — Classify health
50
51	Map tool call count to health level:
52
53	```
54	GREEN (<50 calls) — Healthy, continue normally
55	YELLOW (50-80 calls) — Load only essential files going forward
56	ORANGE (80-120 calls) — Recommend /compact at next logical boundary
57	RED (>120 calls) — Trigger immediate compaction, save state first
58	```
59
60	These thresholds are directional heuristics, not precise limits. Sessions with many large file reads may hit context limits earlier; sessions with mostly Grep/Glob may go longer.
61
62	#### Large-File Adjustment
63
64	Projects with large source files (Python modules often 500-1500 LOC, Java files similarly) consume significantly more context per `Read` call. If the session has read files averaging >500 lines, apply a 0.8x multiplier to all thresholds:
65
66	```
67	Adjusted thresholds (large-file sessions):
68	GREEN (<40 calls) — Healthy, continue normally
69	YELLOW (40-65 calls) — Load only essential files going forward
70	ORANGE (65-100 calls) — Recommend /compact at next logical boundary
71	RED (>100 calls) — Trigger immediate compaction, save state first
72	```
73
74	Detection: count `Read` tool calls that returned >500 lines. If ≥3 such calls → activate large-file thresholds for the remainder of the session.
75
76	### Step 3 — If YELLOW
77
78	Emit advisory to the calling orchestrator:
79
80	> "[X] tool calls. Load only essential files. Avoid reading full files when Grep will do."
81
82	Do NOT trigger compaction yet. Continue execution.
83
84	### Step 4 — If ORANGE
85
86	Emit recommendation to the calling orchestrator:
87
88	> "[X] tool calls. Recommend /compact at next phase boundary (after current module completes)."
89
90	Identify the next safe boundary (end of current loop iteration, end of current file being processed) and flag it.
91
92	Auto-activate Caveman Output Mode (see `references/caveman-mode.md`) — emit `output.density.set` with `mode=caveman, scope=session, source=context-orange`. Reduces output token cost ~75% with no information loss; persists until /compact returns context to GREEN. Manual override (`/caveman` or "stop caveman") always wins.
93
94	### Step 5 — If RED
95
96	Immediately trigger state save via `rune:session-bridge` (Save Mode) before any compaction occurs. If caveman mode was not already active

People who liked this also viewed

6 picks

Type

context-engine

80stars21forks

View on GitHub

How to install context-engine?

rune-kit/rune/context-engine

$curl -o .claude/agents/context-engine.md https://raw.githubusercontent.com/Rune-kit/rune/HEAD/agents/context-engine.md

Installs into the current project.

Summary

Context window management — tracks tool call count, auto-detects when context fills up, triggers state save + compaction. Artifact folding for large outputs (>4000 chars → file + preview).

Files · 1

View on GitHub

agents/context-engine.md

1	# context-engine
2
3	## Purpose
4
5	Context window management for long sessions. Detects when context is approaching limits, triggers smart compaction preserving critical decisions and progress, and coordinates with session-bridge to save state before compaction. Prevents the common failure mode of losing important context mid-workflow.
6
7	### Behavioral Contexts
8
9	Context-engine also manages behavioral mode injection via `contexts/` directory. Three modes are available:
10
11	\| Mode \| File \| When to Use \|
12	\|------\|------\|-------------\|
13	\| `dev` \| `contexts/dev.md` \| Active coding — bias toward action, code-first \|
14	\| `research` \| `contexts/research.md` \| Investigation — read widely, evidence-based \|
15	\| `review` \| `contexts/review.md` \| Code review — systematic, severity-labeled \|
16
17	Mode activation: Orchestrators (cook, team, rescue) can set the active mode by writing to `.rune/active-context.md`. The session-start hook injects the active context file into the session. Mode switches mid-session are supported — the orchestrator updates the file and references the new behavioral rules.
18
19	Default: If no `.rune/active-context.md` exists, no behavioral mode is injected (standard Claude behavior).
20
21	## Triggers
22
23	- Called by `cook` and `team` automatically at context boundaries
24	- Auto-trigger: when tool call count exceeds threshold or context utilization is high
25	- Auto-trigger: before compaction events
26
27	## Calls (outbound)
28
29	# Exception: L3→L3 coordination
30	- `session-bridge` (L3): coordinate state save when context critical
31
32	## Called By (inbound)
33
34	- `cook` (L1): Phase boundaries and when tool count exceeds thresholds
35	- `team` (L1): before parallel workstream dispatch, after merge
36	- `rescue` (L1): between refactoring sessions for state persistence
37	- `context-pack` (L3): when packaging context for sub-agent handoff
38	- `session-bridge` (L3): coordinates with context-engine for compaction timing
39	- `adversary` (L2): (oracle-mode) emit `context.preview` before bundle build to gate token cost
40
41	## Execution
42
43	### Step 1 — Count tool calls
44
45	Count total tool calls made so far in this session. This is the ONLY reliable metric — token usage is not exposed by Claude Code and any estimate will be dangerously inaccurate.
46
47	Do NOT attempt to estimate token percentages. Tool count is a directional proxy, not a precise measurement.
48
49	### Step 2 — Classify health
50
51	Map tool call count to health level:
52
53	```
54	GREEN (<50 calls) — Healthy, continue normally
55	YELLOW (50-80 calls) — Load only essential files going forward
56	ORANGE (80-120 calls) — Recommend /compact at next logical boundary
57	RED (>120 calls) — Trigger immediate compaction, save state first
58	```
59
60	These thresholds are directional heuristics, not precise limits. Sessions with many large file reads may hit context limits earlier; sessions with mostly Grep/Glob may go longer.
61
62	#### Large-File Adjustment
63
64	Projects with large source files (Python modules often 500-1500 LOC, Java files similarly) consume significantly more context per `Read` call. If the session has read files averaging >500 lines, apply a 0.8x multiplier to all thresholds:
65
66	```
67	Adjusted thresholds (large-file sessions):
68	GREEN (<40 calls) — Healthy, continue normally
69	YELLOW (40-65 calls) — Load only essential files going forward
70	ORANGE (65-100 calls) — Recommend /compact at next logical boundary
71	RED (>100 calls) — Trigger immediate compaction, save state first
72	```
73
74	Detection: count `Read` tool calls that returned >500 lines. If ≥3 such calls → activate large-file thresholds for the remainder of the session.
75
76	### Step 3 — If YELLOW
77
78	Emit advisory to the calling orchestrator:
79
80	> "[X] tool calls. Load only essential files. Avoid reading full files when Grep will do."
81
82	Do NOT trigger compaction yet. Continue execution.
83
84	### Step 4 — If ORANGE
85
86	Emit recommendation to the calling orchestrator:
87
88	> "[X] tool calls. Recommend /compact at next phase boundary (after current module completes)."
89
90	Identify the next safe boundary (end of current loop iteration, end of current file being processed) and flag it.
91
92	Auto-activate Caveman Output Mode (see `references/caveman-mode.md`) — emit `output.density.set` with `mode=caveman, scope=session, source=context-orange`. Reduces output token cost ~75% with no information loss; persists until /compact returns context to GREEN. Manual override (`/caveman` or "stop caveman") always wins.
93
94	### Step 5 — If RED
95
96	Immediately trigger state save via `rune:session-bridge` (Save Mode) before any compaction occurs. If caveman mode was not already active

People who liked this also viewed

6 picks

Type