hermes-memory

Name: hermes-memory
Author: AlexAI-MCP

132stars24forks

How to install hermes-memory?

alexai-mcp/hermes-ccc/hermes-memory

$npx -y skills add AlexAI-MCP/hermes-CCC --skill hermes-memory

Installs into the current project.

Summary

Manage durable project memory for Claude Code. Use when prefetching relevant history before work, syncing stable decisions after work, nudging for unsaved lessons, compressing noisy notes, or auditing memory quality.

Files · 1

View on GitHub

SKILL.md

1	# Hermes Memory
2
3	## Purpose
4
5	- Keep long-lived project facts out of ephemeral chat history.
6	- Load only the memory that matters for the current task.
7	- Save reusable lessons instead of repeating the same investigation later.
8	- Prevent memory sprawl by separating durable facts from temporary chatter.
9	- Make project knowledge auditable, editable, and diffable.
10
11	## What Counts As Durable Memory
12
13	- architecture decisions
14	- naming conventions
15	- repository-specific workflows
16	- known failure signatures and fixes
17	- user preferences that change how work should be done
18	- stable environment constraints
19	- reusable commands or validation routines
20	- external integration quirks that repeatedly matter
21
22	## What Does Not Belong In Memory
23
24	- one-off task status
25	- temporary branch names
26	- speculative ideas that were never adopted
27	- verbose transcripts
28	- sensitive secrets or tokens
29	- logs that can be regenerated
30	- transient failures without a confirmed lesson
31
32	## Suggested Layout
33
34	- `memory/MEMORY.md` as the index
35	- one file per topic or subsystem
36	- stable headings inside each memory file
37	- short descriptions in the index so prefetch remains cheap
38
39	## Canonical Headings
40
41	- `## Decisions`
42	- `## Preferences`
43	- `## Tooling Notes`
44	- `## Error Patterns`
45	- `## Validation`
46	- `## Open Questions`
47
48	## Operations
49
50	### Prefetch
51
52	- Use before nontrivial work.
53	- Read the memory index first.
54	- Select only files relevant to the current task.
55	- Summarize the relevant items into a short working context.
56	- Do not dump full memory files into the answer unless asked.
57
58	### Sync
59
60	- Use after completing work with durable lessons.
61	- Review what changed in understanding, not just what files changed.
62	- Merge new facts into the most specific existing file.
63	- Create a new file only when no current topic fits cleanly.
64	- Update `MEMORY.md` if a new file is created.
65
66	### Nudge
67
68	- Use after complex work when valuable lessons exist but should not be saved automatically.
69	- Produce candidates and proposed destinations.
70	- Ask for confirmation when the workflow or environment requires explicit approval.
71	- Prefer small curated memory candidates over bulk transcript dumps.
72
73	### Compress
74
75	- Use when memory files become repetitive or bloated.
76	- Merge duplicate bullets.
77	- Promote stable headings.
78	- Remove stale or contradicted material.
79	- Replace long narratives with short factual bullets.
80
81	### Status
82
83	- Use to audit coverage and hygiene.
84	- Check for missing files referenced by the index.
85	- Check for oversized files that need compression.
86	- Check for duplicate topics across files.
87
88	## Prefetch Procedure
89
90	1. Read the index.
91	2. Match the current task to likely topics.
92	3. Open only the top few relevant files.
93	4. Extract facts that constrain the task.
94	5. Separate hard constraints from softer preferences.
95	6. Return a compact `memory-context` block.
96	7. List what files were checked and which were actually relevant.
97
98	## Sync Procedure
99
100	1. Review the completed work.
101	2. Ask what lesson would help six weeks from now.
102	3. Reject task-local noise.
103	4. Map each accepted lesson to an existing topic file.
104	5. Add or edit concise bullets under stable headings.
105	6. Update the index if the topic is new.
106	7. Report what was saved and where.
107
108	## Compression Rules
109
110	- Keep one canonical phrasing for each stable fact.
111	- Merge near-duplicate fixes into one general error pattern when possible.
112	- Keep commands that were verified.
113	- Remove commands that no longer match the current environment.
114	- Prefer "why plus what" over long narratives.
115
116	## Memory Entry Style
117
118	- Write factual bullets, not paragraphs of storytelling.
119	- Include dates when change over time matters.
120	- Mention affected subsystem, file family, or command family.
121	- Keep each bullet independently useful.
122	- If a rule has exceptions, state them directly.
123
124	## Example Entry
125
126	```markdown
127	## Error Patterns
128
129	- 2026-04-07: PowerShell profile loading can fail under sandboxed runs. Use `login=false` for non-interactive repository inspection commands when possible.
130	```
131
132	## Reporting Format For Prefetch
133
134	````markdown
135	```memory-context
136	- Fact: project uses repo-local AGENTS instructions for coding behavior
137	- Preference: keep edits minimal and verify with targeted checks
138	- Error pattern: sandboxed PowerShell sessions may need profile loading disabled
139	```
140
141	Files checked:
142	- memory/MEMORY.md
143	- memory/tooling.md
144	- memory/repo-workflow.md
145
146	Relevant files:
147	- memory/tooling.md
148	````
149
150	## Reporting Format For Sync
151
152	```markdown
153	Saved memory updates:
154	-

People who liked this also viewed

6 picks

Type

hermes-memory

132stars24forks

View on GitHub

How to install hermes-memory?

alexai-mcp/hermes-ccc/hermes-memory

$npx -y skills add AlexAI-MCP/hermes-CCC --skill hermes-memory

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Hermes Memory
2
3	## Purpose
4
5	- Keep long-lived project facts out of ephemeral chat history.
6	- Load only the memory that matters for the current task.
7	- Save reusable lessons instead of repeating the same investigation later.
8	- Prevent memory sprawl by separating durable facts from temporary chatter.
9	- Make project knowledge auditable, editable, and diffable.
10
11	## What Counts As Durable Memory
12
13	- architecture decisions
14	- naming conventions
15	- repository-specific workflows
16	- known failure signatures and fixes
17	- user preferences that change how work should be done
18	- stable environment constraints
19	- reusable commands or validation routines
20	- external integration quirks that repeatedly matter
21
22	## What Does Not Belong In Memory
23
24	- one-off task status
25	- temporary branch names
26	- speculative ideas that were never adopted
27	- verbose transcripts
28	- sensitive secrets or tokens
29	- logs that can be regenerated
30	- transient failures without a confirmed lesson
31
32	## Suggested Layout
33
34	- `memory/MEMORY.md` as the index
35	- one file per topic or subsystem
36	- stable headings inside each memory file
37	- short descriptions in the index so prefetch remains cheap
38
39	## Canonical Headings
40
41	- `## Decisions`
42	- `## Preferences`
43	- `## Tooling Notes`
44	- `## Error Patterns`
45	- `## Validation`
46	- `## Open Questions`
47
48	## Operations
49
50	### Prefetch
51
52	- Use before nontrivial work.
53	- Read the memory index first.
54	- Select only files relevant to the current task.
55	- Summarize the relevant items into a short working context.
56	- Do not dump full memory files into the answer unless asked.
57
58	### Sync
59
60	- Use after completing work with durable lessons.
61	- Review what changed in understanding, not just what files changed.
62	- Merge new facts into the most specific existing file.
63	- Create a new file only when no current topic fits cleanly.
64	- Update `MEMORY.md` if a new file is created.
65
66	### Nudge
67
68	- Use after complex work when valuable lessons exist but should not be saved automatically.
69	- Produce candidates and proposed destinations.
70	- Ask for confirmation when the workflow or environment requires explicit approval.
71	- Prefer small curated memory candidates over bulk transcript dumps.
72
73	### Compress
74
75	- Use when memory files become repetitive or bloated.
76	- Merge duplicate bullets.
77	- Promote stable headings.
78	- Remove stale or contradicted material.
79	- Replace long narratives with short factual bullets.
80
81	### Status
82
83	- Use to audit coverage and hygiene.
84	- Check for missing files referenced by the index.
85	- Check for oversized files that need compression.
86	- Check for duplicate topics across files.
87
88	## Prefetch Procedure
89
90	1. Read the index.
91	2. Match the current task to likely topics.
92	3. Open only the top few relevant files.
93	4. Extract facts that constrain the task.
94	5. Separate hard constraints from softer preferences.
95	6. Return a compact `memory-context` block.
96	7. List what files were checked and which were actually relevant.
97
98	## Sync Procedure
99
100	1. Review the completed work.
101	2. Ask what lesson would help six weeks from now.
102	3. Reject task-local noise.
103	4. Map each accepted lesson to an existing topic file.
104	5. Add or edit concise bullets under stable headings.
105	6. Update the index if the topic is new.
106	7. Report what was saved and where.
107
108	## Compression Rules
109
110	- Keep one canonical phrasing for each stable fact.
111	- Merge near-duplicate fixes into one general error pattern when possible.
112	- Keep commands that were verified.
113	- Remove commands that no longer match the current environment.
114	- Prefer "why plus what" over long narratives.
115
116	## Memory Entry Style
117
118	- Write factual bullets, not paragraphs of storytelling.
119	- Include dates when change over time matters.
120	- Mention affected subsystem, file family, or command family.
121	- Keep each bullet independently useful.
122	- If a rule has exceptions, state them directly.
123
124	## Example Entry
125
126	```markdown
127	## Error Patterns
128
129	- 2026-04-07: PowerShell profile loading can fail under sandboxed runs. Use `login=false` for non-interactive repository inspection commands when possible.
130	```
131
132	## Reporting Format For Prefetch
133
134	````markdown
135	```memory-context
136	- Fact: project uses repo-local AGENTS instructions for coding behavior
137	- Preference: keep edits minimal and verify with targeted checks
138	- Error pattern: sandboxed PowerShell sessions may need profile loading disabled
139	```
140
141	Files checked:
142	- memory/MEMORY.md
143	- memory/tooling.md
144	- memory/repo-workflow.md
145
146	Relevant files:
147	- memory/tooling.md
148	````
149
150	## Reporting Format For Sync
151
152	```markdown
153	Saved memory updates:
154	-

People who liked this also viewed

6 picks

Type