harness-resume

Name: harness-resume
Author: xwtro0tk1t-cloud

320stars42forks

How to install harness-resume?

xwtro0tk1t-cloud/harness/harness-resume

$npx -y skills add xwtro0tk1t-cloud/harness --skill harness-resume

Installs into the current project.

Summary

--- description: Trigger when user says "harness resume", "resume context", "recover context", "load context", "resume after compact", "I want to continue where I left off", "restore work context". Lightweight same-session context recovery after /compact. Do NOT trigger for: "con

Files · 1

View on GitHub

SKILL.md

1	# Harness Resume — Lightweight Context Recovery
2
3	---
4	description: Trigger when user says "harness resume", "resume context", "recover context", "load context", "resume after compact", "I want to continue where I left off", "restore work context". Lightweight same-session context recovery after /compact. Do NOT trigger for: "continue writing X function" (just code), "help me understand this file" (single file read), bare "resume" without context (too vague), "I'm new here" (use harness handoff instead).
5	---
6
7	## Behavior
8
9	Perform lightweight context recovery for the current project. Core principle: Index-First, Fragment-on-Demand — only load indexes, read fragments on demand.
10	This is for same-session /compact recovery only. For new agent handoff or crash recovery, use `harness handoff`.
11
12	---
13
14	## Steps
15
16	### Step 1: task_plan.md Phase Anchor
17
18	- If `task_plan.md` exists → Read the file
19	- Use regex `^#{1,3}\s.*([Pp]hase)` to locate current Phase heading (covers `#`/`##`/`###`)
20	- Extract ±30 lines around the match (shrink to ±15 if token budget exceeded)
21	- WIP signals: unchecked `- [ ]`, 🚧, WIP, in progress, (doing)
22	- If no `task_plan.md` → skip, output "No task_plan.md found"
23
24	### Step 2: MEMORY.md Best-Match
25
26	Locate the project-specific MEMORY.md using this algorithm:
27
28	```
29	target = cwd.replace('/', '-').lstrip('-')
30	candidates = glob('~/.claude/projects/*/memory/MEMORY.md')
31	matches = [c for c in candidates if c.parents[1].name.lstrip('-') == target]
32	```
33
34	- `len == 1` → Read that file (first 50 lines, index section)
35	- `len == 0` → Skip
36	- `len > 1` → List candidates and ask user to choose
37
38	### Step 3: docs/ INDEX Tree
39
40	- Read `docs/INDEX.md` if it exists
41	- Glob `docs/**/INDEX.md`, read up to 5 files
42	- If no INDEX files → skip
43
44	### Step 4: Output Structured Summary
45
46	```
47	📍 Current Phase: [Phase name / status]
48	⏳ Unfinished: [unchecked items]
49	⏸ Last stopped at: [latest progress entry]
50	🧠 Relevant memory: [MEMORY.md relevant entries]
51	📚 Doc index: [INDEX tree summary]
52
53	Next: Tell me what you want to do, I'll read specific doc fragments on demand
54	```
55
56	---
57
58	## Token Budget
59
60	Target 3-5k tokens:
61	- Phase ±30 lines ≈ 1-2k (shrink to ±15 if budget tight)
62	- MEMORY ≤50 lines ≈ 0.5-1k
63	- INDEX ≤5 files ≈ 0.5-1k
64
65	If exceeded, truncate Phase to ±15 lines first.
66
67	---
68
69	## Empty Rendering
70
71	- If a row field is empty → omit the entire row
72	- If all 4 core rows are empty → print fallback:
73	> No work context found — this may be a new project. Say `harness` to start initialization.
74
75	## Graceful Degradation
76
77	Any file missing → skip that step, output "Step skipped (file not found)" and continue. Never error out.

People who liked this also viewed

6 picks

Type

harness-resume

320stars42forks

View on GitHub

How to install harness-resume?

xwtro0tk1t-cloud/harness/harness-resume

$npx -y skills add xwtro0tk1t-cloud/harness --skill harness-resume

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Harness Resume — Lightweight Context Recovery
2
3	---
4	description: Trigger when user says "harness resume", "resume context", "recover context", "load context", "resume after compact", "I want to continue where I left off", "restore work context". Lightweight same-session context recovery after /compact. Do NOT trigger for: "continue writing X function" (just code), "help me understand this file" (single file read), bare "resume" without context (too vague), "I'm new here" (use harness handoff instead).
5	---
6
7	## Behavior
8
9	Perform lightweight context recovery for the current project. Core principle: Index-First, Fragment-on-Demand — only load indexes, read fragments on demand.
10	This is for same-session /compact recovery only. For new agent handoff or crash recovery, use `harness handoff`.
11
12	---
13
14	## Steps
15
16	### Step 1: task_plan.md Phase Anchor
17
18	- If `task_plan.md` exists → Read the file
19	- Use regex `^#{1,3}\s.*([Pp]hase)` to locate current Phase heading (covers `#`/`##`/`###`)
20	- Extract ±30 lines around the match (shrink to ±15 if token budget exceeded)
21	- WIP signals: unchecked `- [ ]`, 🚧, WIP, in progress, (doing)
22	- If no `task_plan.md` → skip, output "No task_plan.md found"
23
24	### Step 2: MEMORY.md Best-Match
25
26	Locate the project-specific MEMORY.md using this algorithm:
27
28	```
29	target = cwd.replace('/', '-').lstrip('-')
30	candidates = glob('~/.claude/projects/*/memory/MEMORY.md')
31	matches = [c for c in candidates if c.parents[1].name.lstrip('-') == target]
32	```
33
34	- `len == 1` → Read that file (first 50 lines, index section)
35	- `len == 0` → Skip
36	- `len > 1` → List candidates and ask user to choose
37
38	### Step 3: docs/ INDEX Tree
39
40	- Read `docs/INDEX.md` if it exists
41	- Glob `docs/**/INDEX.md`, read up to 5 files
42	- If no INDEX files → skip
43
44	### Step 4: Output Structured Summary
45
46	```
47	📍 Current Phase: [Phase name / status]
48	⏳ Unfinished: [unchecked items]
49	⏸ Last stopped at: [latest progress entry]
50	🧠 Relevant memory: [MEMORY.md relevant entries]
51	📚 Doc index: [INDEX tree summary]
52
53	Next: Tell me what you want to do, I'll read specific doc fragments on demand
54	```
55
56	---
57
58	## Token Budget
59
60	Target 3-5k tokens:
61	- Phase ±30 lines ≈ 1-2k (shrink to ±15 if budget tight)
62	- MEMORY ≤50 lines ≈ 0.5-1k
63	- INDEX ≤5 files ≈ 0.5-1k
64
65	If exceeded, truncate Phase to ±15 lines first.
66
67	---
68
69	## Empty Rendering
70
71	- If a row field is empty → omit the entire row
72	- If all 4 core rows are empty → print fallback:
73	> No work context found — this may be a new project. Say `harness` to start initialization.
74
75	## Graceful Degradation
76
77	Any file missing → skip that step, output "Step skipped (file not found)" and continue. Never error out.

People who liked this also viewed

6 picks

Type