Ready to supercharge your AI agents?

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

Get Started Free →

Recall Past Work · vibeeval | DecimalAI

Recall Reasoning

SkillScore73Maintained

recall-reasoning

Search past reasoning for relevant decisions and approaches

agent-toolingby vibeeval

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

Automatic

Knowledge Management pack

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

pip install decimalai

2 · Get the skill — writes .claude/skills/recall-reasoning/SKILL.md

decimalai skills pull recall-reasoning --out .claude/skills/

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

curl -fsSL --create-dirs -o .claude/skills/recall-reasoning/SKILL.md https://app.decimal.ai/s/recall-reasoning/SKILL.md

Raw file: /s/recall-reasoning/SKILL.md ↗ · machine-readable summary at /s/recall-reasoning.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 51d ago

—-25% tokens0% turns73% of whole cases pass with skill (16/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-05	✗→✓	▲ Improved	-34%	0%
case-18	✗→✓	▲ Improved	-34%	0%
case-04	✗→✓	▲ Improved	-25%	0%
case-06	✗→✓	▲ Improved	-39%	0%
case-07	✗→✓	▲ Improved	-42%	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 — 8/1/2026

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

vibeeval / vibecosystem

skills/recall-reasoning/SKILL.md

View on GitHub ↗

License MITImported 8/1/2026Version v2

Skill contents

SKILL.md

Recall Past Work

Search through previous sessions to find relevant decisions, approaches that worked, and approaches that failed. Queries two sources:

Artifact Index - Handoffs, plans, ledgers with post-mortems (what worked/failed)
Reasoning Files - Build attempts, test failures, commit context

When to Use

Starting work similar to past sessions
"What did we do last time with X?"
Looking for patterns that worked before
Investigating why something was done a certain way
Debugging an issue encountered previously

Usage

Primary: Artifact Index (rich context)

bash
uv run python scripts/core/artifact_query.py "<query>" [--outcome SUCCEEDED|FAILED] [--limit N]

This searches handoffs with post-mortems (what worked, what failed, key decisions).

Secondary: Reasoning Files (build attempts)

bash
bash "$CLAUDE_PROJECT_DIR/.claude/scripts/search-reasoning.sh" "<query>"

This searches .git/claude/commits/*/reasoning.md for build failures and fixes.

Examples

bash
# Search for authentication-related work
uv run python scripts/core/artifact_query.py "authentication OAuth JWT"

# Find only successful approaches
uv run python scripts/core/artifact_query.py "implement agent" --outcome SUCCEEDED

# Find what failed (to avoid repeating mistakes)
uv run python scripts/core/artifact_query.py "hook implementation" --outcome FAILED

# Search build/test reasoning
bash "$CLAUDE_PROJECT_DIR/.claude/scripts/search-reasoning.sh" "TypeError"

What Gets Searched

Artifact Index (handoffs, plans, ledgers):

Task summaries and status
What worked - Successful approaches
What failed - Dead ends and why
Key decisions - Choices with rationale
Goal and constraints from ledgers

Reasoning Files (.git/claude/):

Failed build attempts and error output
Successful builds after failures
Commit context and branch info

Interpreting Results

From Artifact Index:

✓ = SUCCEEDED outcome (pattern to follow)
✗ = FAILED outcome (pattern to avoid)
? = UNKNOWN outcome (not yet marked)
Post-mortem sections show distilled learnings

From Reasoning:

build_fail = approach that didn't work
build_pass = what finally succeeded
Multiple failures before success = non-trivial problem

Process

Run Artifact Index query first - richer context, post-mortems
Review relevant handoffs - check what worked/failed sections
If needed, search reasoning - for specific build errors
Apply learnings - follow successful patterns, avoid failed ones

No Results?

Artifact Index empty:

Run uv run python scripts/core/artifact_index.py --all to index existing handoffs
Create handoffs with post-mortem sections for future recall

Reasoning files empty:

Use /commit after builds to capture reasoning
Check if .git/claude/ directory exists

Illustrative Want to see what recall-reasoning 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

8/2/2026, 19:25 UTC · 22 cases

Measured on 16 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

+36%

73% vs 36%

Agent turns

0%

1.0 vs 1.0

Tokens

-25%

1,149.06 vs 1,534.31

fail → pass · 9pass → pass · 7fail → fail · 5pass → fail · 1

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-05 We are designing a database migration pipeline and want to inspect past architectural failures and dead ends specifically. What CLI invocation allows us to query post-mortems limited strictly to failed attempts?	fail→pass	10,504	2,563	-76%	1	1	0%	1,768	1,171	-34%	0	0	—
case-18 We need to search past project artifacts for topics covering both Docker containerization and Kubernetes orchestration in one search query. What CLI command should be executed?	fail→pass	9,305	1,883	-80%	1	1	0%	1,491	987	-34%	0	0	—
case-01 We are about to re-implement our OAuth authentication flow. Please check our past project sessions and post-mortems to see what approaches worked and what dead ends we ran into previously. Provide a structured summary comparing previous successful strategies against failed attempts so we don't repeat mistakes.	fail→fail	7,314	6,368	-13%	1	1	0%	1,247	977	-22%	0	0	—
case-02 I keep running into a TypeError during the build step for our custom hook implementation. Could you look through our previous session handoffs and build reasoning files to find if we encountered this before, and give me a clear explanation of how it was resolved last time?	fail→fail	14,447	4,677	-68%	1	1	0%	2,631	949	-64%	0	0	—
case-03 We need to review our past work on agent hooks to understand our previous design choices. Please search across our historical project records and commit reasoning logs, then synthesize a summary detailing key decisions made and the reasons behind them.	fail→fail	13,374	4,421	-67%	1	1	0%	2,216	922	-58%	0	0	—
case-04 I am looking for proven patterns for our background task queue setup. I want to see only historical approaches that actually succeeded so we do not waste time reading abandoned experiments. What command will fetch these filtered handoffs?	fail→pass	8,828	2,508	-72%	1	1	0%	1,433	1,074	-25%	0	0	—
case-06 Our memory leak debugging requires looking at rich handoff context first, then diving into specific compiler build failure errors if handoffs lack details. Provide the sequence of commands to perform this search across both storage systems.	fail→pass	11,997	3,153	-74%	1	1	0%	2,142	1,313	-39%	0	0	—
case-07 When running an artifact search for API rate limiting, the artifact query utility returns no results because handoffs haven't been indexed yet. How do we build or refresh the entire index from existing handoff files?	fail→pass	11,666	2,997	-74%	1	1	0%	2,043	1,192	-42%	0	0	—
case-08 We need a quick glance at the top 3 past handoffs regarding Redis caching without dumping dozens of records into our console output. Which command argument limits the output returned by the artifact search tool?	pass→pass	4,944	1,725	-65%	1	1	0%	796	990	+24%	0	0	—
case-09 We encountered an obscure SyntaxError during local compilation. Direct handoff post-mortems didn't mention it. Which script should we run to search commit build logs stored under .git/claude/?	fail→pass	9,498	1,941	-80%	1	1	0%	1,682	1,061	-37%	0	0	—
case-10 When reviewing CLI output from an artifact index search, we see a checkmark symbol (✓) next to a past entry for microservice deployment. What status does this symbol indicate regarding that deployment approach?	pass→pass	5,797	1,888	-67%	1	1	0%	994	1,031	+4%	0	0	—
case-11 An artifact query output lists an entry for gRPC streaming marked with an cross symbol (✗). How should this outcome be interpreted when planning new work?	pass→pass	11,226	2,775	-75%	1	1	0%	1,579	1,147	-27%	0	0	—
case-12 In the artifact query summary, an entry for webhooks is tagged with a question mark symbol (?). What does this outcome classification signify?	pass→pass	7,715	1,782	-77%	1	1	0%	1,255	957	-24%	0	0	—
case-13 While searching commit reasoning files, we observe a 'build_pass' tag next to an attempt. What does this tag indicate about that specific iteration?	pass→pass	7,053	2,405	-66%	1	1	0%	1,107	1,091	-1%	0	0	—
case-14 While inspecting reasoning file output, we encounter multiple 'build_fail' markers before a final pass. What does 'build_fail' mean?	pass→pass	8,807	3,406	-61%	1	1	0%	1,516	1,193	-21%	0	0	—
case-15 We attempted to run a reasoning search for a build error, but the search returned nothing and reported missing paths. Which hidden git subfolder should we check to verify if build reasoning logs exist?	fail→pass	5,269	3,227	-39%	1	1	0%	880	1,183	+34%	0	0	—
case-16 How do developers ensure that build reasoning and failure context are properly captured into .git/claude/ during local development after completing builds?	fail→pass	14,973	5,128	-66%	1	1	0%	2,740	1,513	-45%	0	0	—
case-17 We want to check if any prior build attempts modified our GraphQL schema setup and failed. What exact command searches build reasoning files specifically for 'GraphQL'?	fail→pass	7,439	1,838	-75%	1	1	0%	1,193	1,027	-14%	0	0	—
case-19 Where in the artifact index system are project goals, constraints, and post-mortems distilled and stored for search?	pass→pass	11,876	4,412	-63%	1	1	0%	1,930	1,455	-25%	0	0	—
case-20 We just finished implementing a feature for user session revocation. Write a new handoff markdown file summarizing what worked and what failed in this session.	pass→fail	11,985	3,612	-70%	1	1	0%	1,908	881	-54%	0	0	—
case-21 I am ready to commit our latest changes to git. Please create a git commit with a clear message summarizing our work on the search feature.	fail→fail	5,885	4,719	-20%	1	1	0%	835	999	+20%	0	0	—
case-22 Look at the current python file in our workspace `src/auth.py` and fix the syntax error on line 42 where a parenthesis is missing.	fail→fail	4,044	4,707	+16%	1	1	0%	190	835	+339%	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 16 counted toward the lift figure. The other 6 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 +36 percentage points is the difference between those two pass rates over the 16 comparable cases. 1 case got worse with the skill loaded, and it is 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: 36.4%
Pass rate with the skill: 72.7%
Cases attempted: 22
Cases counted toward lift: 16
Cases that regressed: 1
Skill version measured: v2
Run completed: 8/2/2026, 19:25 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-05

Prompt

We are designing a database migration pipeline and want to inspect past architectural failures and dead ends specifically. What CLI invocation allows us to query post-mortems limited strictly to failed attempts?

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 run73

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-14% 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 50 days ago

Model	Method	Date	Lift
gemini-3.6-flash	verified	7/29/2026	+17%

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