Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Session Memory · a5c-ai | DecimalAI

a5c-ai/session-memory

SkillScore50Maintained

Mandatory memory persistence system across session resets using three markdown surfaces in .claude/cc10x/. Iron law - every workflow must load at start and update at end.

agent-toolingby a5c-ai

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/a5c-ai-session-memory/SKILL.md

decimalai skills pull a5c-ai-session-memory --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/a5c-ai-session-memory/SKILL.md https://app.decimal.ai/s/a5c-ai-session-memory/SKILL.md

Raw file: /s/a5c-ai-session-memory/SKILL.md ↗ · machine-readable summary at /s/a5c-ai-session-memory.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 55d ago

—-36% tokens0% turns50% of whole cases pass with skill (11/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	-11%	0%
case-05	✗→✓	▲ Improved	-59%	0%
case-06	✗→✓	▲ Improved	-45%	0%
case-07	✗→✓	▲ Improved	-47%	0%
case-08	✗→✓	▲ Improved	-58%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 7/31/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

a5c-ai / babysitter

library/methodologies/cc10x/skills/session-memory/SKILL.md

View on GitHub ↗

License MITImported 7/31/2026Version v1

Skill contents

SKILL.md

Iron Law

EVERY WORKFLOW MUST:

LOAD memory at START (and before key decisions)
UPDATE memory at END (and after learnings/decisions)

Stable Edit Anchors

Safe section headers for Edit operations:

activeContext: ## Recent Changes, ## Learnings, ## References
patterns: ## Common Gotchas, ## Project SKILL_HINTS
progress: ## Completed, ## Verification

Read-Edit-Verify Pattern

Read file
Verify anchor exists
Edit with exact old_string
Read back to confirm

Tool Rules

Use Write() for NEW files (permission-free)
Use Edit() for EXISTING files (permission-free)
Never use Write() to overwrite existing files
Never compound commands (mkdir && cat)

When to Use

At the start of every CC10X workflow (load)
At the end of every CC10X workflow (update)
Before making key decisions (check patterns)
After discovering learnings or gotchas (persist)

Agents Used

All CC10X agents use this skill. The cc10x-router manages load/update lifecycle.

Illustrative Want to see what a5c-ai/session-memory would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

7/31/2026, 22:45 UTC · 22 cases

Measured on 18 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+41%

50% vs 9%

Agent turns

1.0 vs 1.0

Tokens

-36%

824.06 vs 1,285.33

fail → pass · 9pass → pass · 2fail → fail · 11

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I'm starting a new multi-step feature development task on this project. Before we make any key architectural choices, please read through our project session memory files to get up to speed on context and past gotchas, and then create a new progress tracking log for today's session to record what we complete.	fail→fail	2,688	4,614	+72%	1	1	0%	273	527	+93%	0	0	—
case-02 We just finished debugging a tricky issue with our database pool connection dropping under load. Please update our existing session memory file with these new findings and gotchas so future agent runs won't repeat this mistake, ensuring you verify the existing section anchors and check your changes afterward.	fail→fail	3,071	5,502	+79%	1	1	0%	270	648	+140%	0	0	—
case-03 We need to pause here and log our current progress before moving to the next refactoring phase. Could you inspect our active session context, check if there are known patterns we should avoid, update the session notes file safely without overwriting it completely, and confirm the updates are reflected?	fail→fail	1,722	4,980	+189%	1	1	0%	301	726	+141%	0	0	—
case-04 We need to set up a brand new session memory file named `.context/patterns.md` to track project hints. There is no existing file at that path. How should this new file be created?	fail→pass	5,456	3,026	-45%	1	1	0%	902	801	-11%	0	0	—
case-05 We have an existing `.context/activeContext.md` file with extensive history. I want to add a quick note about our recent refactoring. A developer suggested calling `Write()` on the path with the updated full text. Is that acceptable or should a different approach be used?	fail→pass	10,510	2,928	-72%	1	1	0%	1,844	762	-59%	0	0	—
case-06 When setting up session tracking directories and files, a script suggests running `mkdir -p .context && cat << 'EOF' > .context/progress.md`. How should tool operations be structured for this task?	fail→pass	9,293	3,936	-58%	1	1	0%	1,742	956	-45%	0	0	—
case-07 We just learned that the API returns 429 errors if parallel requests exceed 5 worker threads. We need to append this finding into `activeContext.md`. Which section header anchor should be targeted in this file?	fail→pass	6,623	2,084	-69%	1	1	0%	1,174	625	-47%	0	0	—
case-08 Our build breaks if Node 18 is used instead of Node 20. We want to record this trap in `patterns.md`. What specific section anchor in `patterns.md` should hold this information?	fail→pass	8,327	1,956	-77%	1	1	0%	1,434	606	-58%	0	0	—
case-09 We have successfully finished task #402 (JWT authentication middleware migration). We need to mark this in `progress.md`. Under which section header anchor should completed work items be placed?	pass→pass	6,346	1,875	-70%	1	1	0%	1,121	534	-52%	0	0	—
case-10 We need to add a link to external API specifications into `activeContext.md`. Which section header anchor in `activeContext.md` is designated for external links?	fail→pass	6,521	1,752	-73%	1	1	0%	1,215	546	-55%	0	0	—
case-11 We ran end-to-end integration tests and saved test logs. We need to update `progress.md` to show that the migration passed validation. Which section header anchor in `progress.md` should receive these test results?	fail→pass	7,604	3,028	-60%	1	1	0%	1,373	770	-44%	0	0	—
case-12 I am about to edit `patterns.md` to update our project hints. Walk me through the step-by-step sequence required before and after applying the file modification.	fail→pass	11,213	7,958	-29%	1	1	0%	1,682	1,836	+9%	0	0	—
case-13 When should session memory be loaded and updated during an agent task workflow execution?	fail→fail	12,994	2,439	-81%	1	1	0%	2,238	707	-68%	0	0	—
case-14 We need to configure our repository's `.git/hooks/commit-msg` script to enforce local Git commit message syntax such as `feat: description`. How should this hook script validate commit messages?	fail→fail	10,856	4,472	-59%	1	1	0%	2,557	605	-76%	0	0	—
case-15 We are configuring Python's `logging` module in `server.py` to write runtime application logs to `/var/log/app.log` with a custom formatter `%(asctime)s - %(levelname)s - %(message)s`. How should the Logger instance be initialized?	fail→fail	9,663	7,649	-21%	1	1	0%	1,602	1,784	+11%	0	0	—
case-16 We need to write a SQL migration script `V2__create_user_sessions_table.sql` to store user web session tokens (`session_id VARCHAR PRIMARY KEY`, `user_id INT`, `expires_at TIMESTAMP`). What SQL statement should be used?	fail→fail	5,199	6,546	+26%	1	1	0%	837	1,021	+22%	0	0	—
case-17 We want to record specific prompt shortcuts and agent operational tips in `patterns.md`. Which anchor section header is intended for these operational hints?	fail→pass	6,345	2,294	-64%	1	1	0%	1,025	666	-35%	0	0	—
case-18 We refactored the database connection pooling module 10 minutes ago. We want to record this recent code modification in `activeContext.md`. Which section anchor header should be used?	fail→fail	2,871	1,807	-37%	1	1	0%	480	545	+14%	0	0	—
case-19 When modifying an existing session memory document using an edit tool, what requirement applies to the `old_string` parameter?	fail→fail	3,881	1,793	-54%	1	1	0%	610	548	-10%	0	0	—
case-20 We are about to select a state management library for our frontend refactor. Before finalizing this architectural choice, what session memory step must be performed?	pass→pass	10,614	2,621	-75%	1	1	0%	1,579	717	-55%	0	0	—
case-21 During testing, we discovered that Redis key eviction causes silent token loss under high memory pressure. What action must be taken regarding session memory immediately after discovering this gotcha?	fail→fail	5,743	3,006	-48%	1	1	0%	938	838	-11%	0	0	—
case-22 When choosing between file manipulation tools, what rule specifies tool selection based on whether the file already exists?	fail→fail	7,965	1,891	-76%	1	1	0%	1,340	571	-57%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +41 percentage points is the difference between those two pass rates over the 18 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 9.1%
Pass rate with the skill: 50%
Cases attempted: 22
Cases counted toward lift: 18
Cases that regressed: 2
Skill version measured: v1
Run completed: 7/31/2026, 22:45 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

We need to set up a brand new session memory file named `.context/patterns.md` to track project hints. There is no existing file at that path. How should this new file be created?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run50

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns-54% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 55 days ago

Other measured skills in the registry, with their headline benchmark lift.