Session Context Extractor

What This Skill Does

After each session, this skill extracts and stores:

• Decisions — what you chose, why, what you rejected, when to reconsider
• Errors — specific error messages, what you tried, what worked or failed
• Preferences — casual mentions like "I hate that tool" or "I prefer this approach"
• Contacts — who you've talked to and what you discussed

These get stored in a queryable vault (memory/context-vault/) so you can answer questions like:

• "What did we decide about OCI vs Azure?"
• "What errors have we hit with Proxmox?"
• "What does Jason prefer for logging tools?"
• "What was the last thing we tried for [system]?"

When to Use This Skill

Use this skill when:

• You make a significant decision and want the reasoning preserved
• You hit an error and want the specific error messages + attempts logged
• You mention a preference (even casually) and want it remembered
• You want to query what you've learned about a system or topic across multiple sessions

How It Works

Automatic Extraction

When you tell me "Use the session-context-extractor skill," I will:

1. Identify decisions made this session — what was chosen, alternatives considered, reasoning
2. Extract errors and troubleshooting — specific error messages, attempts made, resolution
3. Capture preferences — things you mentioned liking/disliking
4. Track contacts — people mentioned and context

The Vault Structure

Your facts are stored at memory/context-vault/ as:

memory/context-vault/
  vault.db                     # SQLite index (queryable, machine-readable)
  decision/                    # Individual decision files
    2026-04-26-oci-vs-azure.md
  error/                       # Error history by incident
    2026-04-26-proxmox-fs-corruption.md
  preference/                  # Your stated preferences
    2026-04-20-logging-tool-preference.md
  contact/                     # People and discussions
    2026-04-15-delta-utilities-conversation.md
  INDEX.md                     # Human-readable summary

How to Query

Ask me questions like:

• "What did we decide about OCI?"
• "Show me all errors with Proxmox"
• "What are Jason's preferences?"
• "Who have we talked to about the homelab?"

I'll search the vault and give you exact reasoning + timestamps + context.

Preferences & Verification

For casual preferences (e.g., "I hate Splunk's UI"), I'll capture them as unverified and ask later:

"You mentioned disliking Splunk's UI — want me to log that as a preference?"

This prevents me from over-inferring. You stay in control.

Stale Decisions

Decisions older than 90 days get flagged as potentially stale. I'll note when something was decided and whether it might need revisiting.

Instructions for Using This Skill

During a Session (How I Should Handle This)

If you make a decision:

You: "I think we should focus on OCI certs first, not Azure."
Me: [extract and confirm] "Got it — decision captured: priority OCI over Azure (deadline pressure + Delta uses OCI). Should I also note when this should be revisited?"
You: "Yeah, revisit if Delta switches to Azure."
Me: [store with revisit trigger]

If you hit an error:

You: "Proxmox dropped into emergency mode again."
Me: [extract details from your description] "Error captured: emergency mode at boot. Details stored: /dev/mapper/pve-root fsck failed, attempted recovery mode — unsuccessful. Reinstall planned as resolution."

If you mention a preference:

You: "I really hate managing Splunk dashboards manually."
Me: [flag for verification] "Noted — potential preference: dislike of manual Splunk dashboard management. Confirm?"
You: "Yeah, remember that."
Me: [marks as verified]

Querying the Vault

Later, you can ask me:

You: "What did we decide about the Proxmox rebuild?"
Me: [searches vault, returns from decision + error logs] "Decision from April 26: reinstall Proxmox over the existing disk to preserve pve-data storage. Reason: filesystem corruption from repeated ungraceful shutdowns. Previous attempts (fsck journal repair) left the system in emergency mode. Revisit trigger: after 32GB RAM upgrade is complete."

Manual Updates

You can also ask me to update the vault directly:

• "Mark preference about Splunk as verified"
• "Add a lesson learned from the Proxmox rebuild: take VM snapshots before major changes"
• "Update the OCI decision: revisit in Q3 2026"

Examples

Decision Extraction Example

Input: You tell me you've decided to prioritize OCI Foundations Associate cert over AZ-104.

What Gets Stored:

Title: Prioritize OCI Foundations Associate over AZ-104
Type: Decision
Date: 2026-04-26
Status: Verified

Choice: OCI Foundations (1Z0-1085) is now priority cert
Reasoning: Delta Utilities uses OCI. AZ-900 already passed. OCI focus aligns with new job.
Constraints: Limited study time, working full-time
Rejected: Parallel study (too much cognitive load)
Revisit Trigger: After OCI is passed, reassess AZ-104 timing

Error Extraction Example

Input: You describe the Proxmox filesystem corruption incident.

What Gets Stored:

Title: Proxmox Filesystem Corruption — Emergency Mode
Type: Error
Date: 2026-04-24
System: Proxmox (Dell OptiPlex 7070)
Status: Unverified (waiting for more details)

Affected Component: /dev/mapper/pve-root
Error Messages:
  - Filesystem corruption detected at boot
  - fsck journal repair failed
  - Dropped to emergency mode, boot stalled

Attempts:
  1. fsck /dev/mapper/pve-root — FAILURE (journal repair incomplete)
  2. Boot recovery mode — FAILURE (emergency mode persisted)

Resolution: Planned Proxmox reinstall over existing disk to preserve pve-data storage

Lessons:
  - Take VM snapshots (202/203/204) before major changes
  - Track repeated ungraceful shutdowns as early warning
  - Have recovery steps documented in advance

Preference Extraction Example

Input: You casually mention "I hate dealing with Splunk's UI."

What Gets Stored (unverified):

Title: Tool Preference — Splunk UI
Type: Preference
Date: 2026-04-26
Topic: Logging tools
Preference: Dislike of Splunk's UI (prefer simpler alternatives)
Strength: Strong
Context: Mentioned in context of dashboard management frustration
Status: Unverified — awaiting confirmation

Later: You: "Yeah, remember that." Me: [marks as verified, can now act on it]

Common Patterns I Should Use

Confirm High-Stakes Preferences

If you mention something casually that sounds important, I confirm before storing:

You: "I'd really prefer not to SSH into that server."
Me: "Noted — should I mark 'avoid SSH to [server]' as a standing preference?"

Prompt for Revisit Triggers

For decisions, I'll ask:

Me: "When should we revisit this decision? (e.g., 'in 6 months', 'if X changes', 'after Y is complete')"

Group Related Facts

If you hit an error that's related to a previous decision, I'll note the connection:

Error (current): Proxmox filesystem corruption
Related Decision: Decided to postpone VM snapshots until after April work
Connection: Snapshots would have prevented this incident

Limitations

• I can't extract facts from before this skill is enabled. Existing MEMORY.md stays as-is; vault captures new context going forward.
• Preferences need explicit confirmation before I'll act on them. I won't infer strong opinions.
• Errors need you to describe what happened; I'll parse your descriptions but won't auto-collect system logs (yet).

Querying the Vault (For Advanced Users)

If you want to query directly, the vault is at memory/context-vault/:

# List recent facts
npm run list

# Search for decisions about OCI
npm run query "OCI decisions"

# Show unverified preferences
npm run unverified

# Export to JSON
npm run export json

See README.md for full command reference.

Integration with MEMORY.md

The vault is separate from MEMORY.md, but complementary:

• Vault = machine-readable, queryable, auto-populated, structured by type
• MEMORY.md = human-readable narrative, manually curated, executive summary

During weekly hygiene, I'll synthesize important vault entries into MEMORY.md sections so both stay in sync.

What This Replaces

This skill doesn't replace MEMORY.md. It enhances it by:

• Capturing context that usually gets lost
• Making facts queryable instead of requiring manual reading
• Tracking why decisions were made, not just what
• Flagging stale decisions
• Preserving error history for troubleshooting

Think of it as MEMORY.md + an index + a reasoning layer.