Is session-learn safe to install?

SaferSkills independently audited session-learn (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.

CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

100

35.0 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 0 flagged

Securityscore 100 · 0 findings

✓—All security checks passedNo findings in this category for the latest scan.pass

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

skills/session-learn/SKILL.md · 1 file

skills/session-learn/SKILL.md4.7 KB · Markdown

name: session-learn

description: Extract reusable knowledge from a work session and save concepts, entities, corrections, patterns, ideas, decisions, and gaps to the wiki. Use when ending a session or when the user says to extract knowledge.

version: "1.1"

updated: "2026-05-22"

assumes: "The current session contains reusable learning worth writing back to durable storage."

conflicts_with: "Do not rewrite immutable source notes; use wiki-ingest for raw source capture and append logs instead."

Session Learn

Extract knowledge signals from current session and persist to wiki.

Usage Template

Prompt

Use session-learn on this conversation. Extract reusable concepts, decisions, corrections, patterns, ideas, and gaps into my wiki.

Use Case

Ending a substantive work session without losing reusable learning to chat history.

Expected Result

The agent creates or updates notes for concepts, decisions, corrections, and reusable patterns.

Output Example

A session learning note with concepts, decisions, corrections, reusable patterns, and open gaps.

Verification Case

The output lists saved files and separates confirmed learning from open questions.

Verified Effect

Useful session knowledge is preserved as durable notes instead of disappearing into chat history.

Success Metrics

Extracts signals into durable notes or explicitly states that no durable signal met the save threshold.
Saved items are classified as concept, entity, correction, pattern, idea, decision, or gap.
Final summary lists saved files and open questions separately.

When to Use

"Extract this session" or "save what we learned"
After substantive work (≥5 tool calls)
Session-end review routine
User corrects agent's understanding
Auto-trigger after cognitive-compile, wiki-ingest, deep-research

7 Knowledge Signals

Resolve destination paths from system/config.md when available. If no config exists, use the default wiki and system paths.

#	Signal	Action
1	New Concept	Create `CONCEPTS_DIR/[name].md` status:seed
2	New Entity	Create `ENTITIES_DIR/[name].md` type:entity
3	Correction	Update page + annotate + log
4	Reusable Pattern	Create/update SOP page
5	New Idea	Save to `CREATIVITY_DIR/ideas/` or `OUTPUTS_DIR/`
6	Major Decision	Record to `DECISIONS_DIR/` with rationale
7	Knowledge Gap	Append to review queue

Closure Protocol ⭐

Every significant output must trigger a closure cycle to prevent value leakage.

3-Step Cycle

FORMAT → LINK → LOG
Convert to wiki form → Cross-reference → Record to log.md

Triggers

Source	Trigger	Action
Cognitive compile	Complete	Extract 3 judgments → wiki/outputs/ + update entities
Wiki ingest	Source ingested	Auto-create entities + update overview + central index
Deep research	Report delivered	Save findings to concept pages + update sources
Behavior design	SOP created	Link SOP to goals + habits
Code project	Feature complete	Save decisions + lessons
User correction	Corrected	Log + propagate to related pages

Conflict Resolution

Type	Resolution
Direct contradiction	Keep both, mark `> [!conflict] YYYY-MM-DD`
Outdated	Add `> [!updated] YYYY-MM-DD`, move old to timeline
Different perspective	Add as alternative view with source
Low confidence	`status: seed`, `evidence_level: single-source`

Knowledge Gaps

Gap	Action
No sources	Add to review queue
Only 1 source	Mark `knowledge_stage: single-source`
Contradictions	Add conflict banner
Undefined concept	Create stub with `status: seed`

Quality Rules

Don't over-extract: <5 tool calls → skip
Deduplicate first: search before creating
Closure always: every output → format → link → log
Confidence markers: new pages get status: seed
Conflict honesty: never overwrite without annotation
Log everything: append to LOG_FILE

Execution

1. Check depth (≥5 tool calls?)
2. Scan for 7 signal types
3. For each: search wiki → update/create
4. Execute closure cycle
5. Check gaps → append to review queue
6. Batch write (parallel)
7. Update overview + central index
8. Append to `LOG_FILE`

Quality Gates

[ ] Session qualified (≥5 tool calls or user-initiated)
[ ] No duplicates created
[ ] New pages: status: seed + ≥2 wikilinks
[ ] Closure cycle executed for all outputs
[ ] Conflicts annotated (not silently overwritten)
[ ] Gaps recorded in review queue
[ ] Log updated

session-learn— agent skill

Is session-learn safe to install?

Session Learn

Usage Template

Success Metrics

When to Use

7 Knowledge Signals

Closure Protocol ⭐

3-Step Cycle

Triggers

Conflict Resolution

Knowledge Gaps

Quality Rules

Execution

Quality Gates

Audit the pieces. Scan the whole. Decide.