Memento Skill
Overview
Memento Skill treats research memory as a controlled recall problem. External
records help an agent continue work across sessions, but they can also pollute
the next decision if stale notes, weak evidence, or misleading summaries are
read as current facts.
The goal is not to remember everything. The goal is to keep the current decision
surface clear: facts in one ledger, contrasts in another, beliefs in a third,
and stale fragments behind an archive index until there is a real reason to
recall them.
Default stance: optimize for fast continuation, not exhaustive replay. A good
tracker lets a fresh agent recover the current decision surface quickly without
re-reading the whole experimental past.
Use this skill when
- The user wants an experiment tracker, ablation plan, run table, or research log.
- The project has many runs and needs a global landscape rather than ad hoc notes.
- The user wants help picking the next experiment by information gain, not intuition alone.
- The user wants to compare predictions against outcomes and update beliefs systematically.
- The user wants to prevent old context, archived findings, or memory fragments
from becoming accidental instructions.
Core rules
- Maintain both views at once: the global landscape and the next local gradient.
- Separate active decision support from archival completeness.
- Add a contrast before launching a run.
- Change one primary axis per contrast; keep the rest controlled enough to interpret.
- Prefer experiments whose outcome would most change the roadmap.
- Negative results are useful if the prediction was explicit and the control was clean.
- The worst result is a bad contrast that teaches nothing.
- If an old experiment no longer changes the next decision, demote it out of the
default working set.
- Do not let every note become a tattoo: a fragment must be classified,
checked, and placed in the right memory layer before it drives action.
Required working set
runs.csv: factual ledger, one row per run.
contrasts.csv: reasoning ledger, one row per controlled comparison.
hypotheses.md: belief ledger, one section per claim and update rule.
If these files do not exist, create them. Use the starter schemas in references/TEMPLATES.md.
Preferred layered layout
For medium or large research programs, prefer this layout:
CURRENT_STATE.md: the minimum entry point for a fresh agent. Keep it short.
ACTIVE_TRACKER.md or ACTIVE_TRACKER.csv: only experiments that still have
decision gradient for the current roadmap.
EVIDENCE_LOG.md: compressed statement of what the project currently believes
and which few experiments justify that belief.
runs.csv, contrasts.csv, hypotheses.md: the full ledger.
archive/: experiments and tracker snapshots that should not be read by
default.
- optional
ARCHIVE_INDEX.md: a searchable index for cold-memory recall.
- optional archive cards: one short note per archived cluster explaining when it
should ever be revisited.
- optional
RECALL_NOTES/: small audit trail for archive recalls that actually
happened.
Think of these as memory layers:
- hot path:
CURRENT_STATE.md, ACTIVE_TRACKER.*, EVIDENCE_LOG.md
- full ledger:
runs.csv, contrasts.csv, hypotheses.md
- cold memory:
archive/, ARCHIVE_INDEX.md, archive cards, RECALL_NOTES/
Compatibility rule:
- Do not break projects that only have
runs.csv, contrasts.csv,
hypotheses.md.
- When the tracker has become too large, introduce the layered files above and
make them the default reading path.
- Do not require archive files for projects that do not need cold-memory recall
yet.
Archive recall model
Treat archive as a cold-memory layer, not as dead storage and not as part of the
default read path.
Rules:
- Do not read archive by default.
- Only recall archive when the current problem justifies it.
- Recall should begin from an index or archive card, not from blind directory
traversal.
- Recall is for extracting reusable assets, not for replaying the entire old
branch.
- If recall finds a still-useful result, promote that result back into the hot
path (
CURRENT_STATE.md, ACTIVE_TRACKER.*, EVIDENCE_LOG.md) rather than
leaving it buried.
Recommended recall triggers:
- current mainline has stalled or failed repeatedly
- current question is structurally similar to an archived question
- a new result conflicts with the active project story
- a high-value anomaly appears and active evidence does not explain it
- the user explicitly asks for historical excavation
Recommended recall budget:
- first read
ARCHIVE_INDEX.md or the relevant archive card
- select at most 1-2 archive clusters
- read at most 1-3 source files per cluster before deciding whether to stop
- write a short recall note if the recall materially informs the next step
Workflow
- Read the smallest authoritative entry point first: prefer
CURRENT_STATE.md,
then ACTIVE_TRACKER.*, then EVIDENCE_LOG.md, then the full ledger only as
needed.
- Identify the active questions, hypotheses, and unresolved decision points.
- Mark old experiments as one of:
active: still changes current decisions
reference: useful supporting evidence but not part of the default working set
archived: historical only unless a specific question reopens
- Propose candidate contrasts and rank them by expected information gain.
- Write the prediction before execution: direction, minimum meaningful delta,
and reason.
- Record observed facts in
runs.csv, not interpretation.
- Update
contrasts.csv with actual deltas, information gain, and next
action.
- Update
hypotheses.md based on the gap between expected and observed
results.
- If archive recall is triggered, consult
ARCHIVE_INDEX.md or archive cards,
do a bounded read, and summarize reusable findings in a recall note.
- Promote any recalled high-value finding into the hot path instead of keeping
it archive-only.
- Refresh
CURRENT_STATE.md and ACTIVE_TRACKER.* so a new agent can resume
without replaying the whole ledger.
- If tracker files exist, run
python <path-to-this-skill>/scripts/validate_tracker.py <tracker_dir>
after editing them. Resolve the script path relative to this SKILL.md.
Deliverables
When using this skill, produce:
- A concise landscape summary
- The most informative next contrasts, ranked
- The control requirements for each contrast
- The predicted gradient or decision impact for each contrast
- Explicit hypothesis updates after new evidence
- A statement of what moved into the active set, what stayed as reference, and
what was archived out of the default reading path
- When archive recall happened: why recall was triggered, what was read, what
reusable asset was found, and whether it changed the plan
Quality bar
runs.csv should store facts and quality flags, not conclusions.
contrasts.csv should make it obvious what changed, what stayed controlled, what was predicted, and what was learned.
hypotheses.md should state what would change the belief, not just the belief itself.
- If a contrast is weakly controlled, mark it inconclusive instead of pretending it resolved the question.
CURRENT_STATE.md should be readable in one short sitting.
ACTIVE_TRACKER.* should contain only the experiments with nontrivial current
decision gradient.
- Historical completeness belongs in the ledger and archive, not in the default
agent entry point.
- Archive should be indexed by problem value, not only by date or snapshot name.
- Useful archived findings should be promotable back into the hot path with
minimal rewriting.
See references/REFERENCE.md for the full rubric and references/TEMPLATES.md for starter files and examples.
1---2name: memento-skill3description: Controlled recall for research memory, experiment tracking, evidence ledgers, hypothesis updates, archival recall, and next high-information experimental contrasts.4license: MIT5category: research-data-analysis6section: research-data-analysis7collection: community8short_description: Controlled recall for research memory and experiment tracking.9metadata:10 source: https://skillmd.com/skills/waterdrop26651/memento-skill11 author: waterdrop2665112 license: MIT13 version: 1.0.014 categories:15 - research-data-analysis16 - research17 - agent-memory18 - experiment-tracking19 tags:20 - agent-skills21 - skill-md22 - codex23 - claude-code24 - external-memory25 - ai-memory26 - hypotheses27 - experiments28---2930# Memento Skill3132## Overview3334Memento Skill treats research memory as a controlled recall problem. External35records help an agent continue work across sessions, but they can also pollute36the next decision if stale notes, weak evidence, or misleading summaries are37read as current facts.3839The goal is not to remember everything. The goal is to keep the current decision40surface clear: facts in one ledger, contrasts in another, beliefs in a third,41and stale fragments behind an archive index until there is a real reason to42recall them.4344Default stance: optimize for fast continuation, not exhaustive replay. A good45tracker lets a fresh agent recover the current decision surface quickly without46re-reading the whole experimental past.4748## Use this skill when4950- The user wants an experiment tracker, ablation plan, run table, or research log.51- The project has many runs and needs a global landscape rather than ad hoc notes.52- The user wants help picking the next experiment by information gain, not intuition alone.53- The user wants to compare predictions against outcomes and update beliefs systematically.54- The user wants to prevent old context, archived findings, or memory fragments55 from becoming accidental instructions.5657## Core rules5859- Maintain both views at once: the global landscape and the next local gradient.60- Separate active decision support from archival completeness.61- Add a contrast before launching a run.62- Change one primary axis per contrast; keep the rest controlled enough to interpret.63- Prefer experiments whose outcome would most change the roadmap.64- Negative results are useful if the prediction was explicit and the control was clean.65- The worst result is a bad contrast that teaches nothing.66- If an old experiment no longer changes the next decision, demote it out of the67 default working set.68- Do not let every note become a tattoo: a fragment must be classified,69 checked, and placed in the right memory layer before it drives action.7071## Required working set7273- `runs.csv`: factual ledger, one row per run.74- `contrasts.csv`: reasoning ledger, one row per controlled comparison.75- `hypotheses.md`: belief ledger, one section per claim and update rule.7677If these files do not exist, create them. Use the starter schemas in [references/TEMPLATES.md](references/TEMPLATES.md).7879## Preferred layered layout8081For medium or large research programs, prefer this layout:8283- `CURRENT_STATE.md`: the minimum entry point for a fresh agent. Keep it short.84- `ACTIVE_TRACKER.md` or `ACTIVE_TRACKER.csv`: only experiments that still have85 decision gradient for the current roadmap.86- `EVIDENCE_LOG.md`: compressed statement of what the project currently believes87 and which few experiments justify that belief.88- `runs.csv`, `contrasts.csv`, `hypotheses.md`: the full ledger.89- `archive/`: experiments and tracker snapshots that should not be read by90 default.91- optional `ARCHIVE_INDEX.md`: a searchable index for cold-memory recall.92- optional archive cards: one short note per archived cluster explaining when it93 should ever be revisited.94- optional `RECALL_NOTES/`: small audit trail for archive recalls that actually95 happened.9697Think of these as memory layers:9899- hot path: `CURRENT_STATE.md`, `ACTIVE_TRACKER.*`, `EVIDENCE_LOG.md`100- full ledger: `runs.csv`, `contrasts.csv`, `hypotheses.md`101- cold memory: `archive/`, `ARCHIVE_INDEX.md`, archive cards, `RECALL_NOTES/`102103Compatibility rule:104105- Do not break projects that only have `runs.csv`, `contrasts.csv`,106 `hypotheses.md`.107- When the tracker has become too large, introduce the layered files above and108 make them the default reading path.109- Do not require archive files for projects that do not need cold-memory recall110 yet.111112## Archive recall model113114Treat archive as a cold-memory layer, not as dead storage and not as part of the115default read path.116117Rules:118119- Do not read archive by default.120- Only recall archive when the current problem justifies it.121- Recall should begin from an index or archive card, not from blind directory122 traversal.123- Recall is for extracting reusable assets, not for replaying the entire old124 branch.125- If recall finds a still-useful result, promote that result back into the hot126 path (`CURRENT_STATE.md`, `ACTIVE_TRACKER.*`, `EVIDENCE_LOG.md`) rather than127 leaving it buried.128129Recommended recall triggers:130131- current mainline has stalled or failed repeatedly132- current question is structurally similar to an archived question133- a new result conflicts with the active project story134- a high-value anomaly appears and active evidence does not explain it135- the user explicitly asks for historical excavation136137Recommended recall budget:138139- first read `ARCHIVE_INDEX.md` or the relevant archive card140- select at most 1-2 archive clusters141- read at most 1-3 source files per cluster before deciding whether to stop142- write a short recall note if the recall materially informs the next step143144## Workflow1451461. Read the smallest authoritative entry point first: prefer `CURRENT_STATE.md`,147 then `ACTIVE_TRACKER.*`, then `EVIDENCE_LOG.md`, then the full ledger only as148 needed.1492. Identify the active questions, hypotheses, and unresolved decision points.1503. Mark old experiments as one of:151 - `active`: still changes current decisions152 - `reference`: useful supporting evidence but not part of the default working set153 - `archived`: historical only unless a specific question reopens1544. Propose candidate contrasts and rank them by expected information gain.1555. Write the prediction before execution: direction, minimum meaningful delta,156 and reason.1576. Record observed facts in `runs.csv`, not interpretation.1587. Update `contrasts.csv` with actual deltas, information gain, and next159 action.1608. Update `hypotheses.md` based on the gap between expected and observed161 results.1629. If archive recall is triggered, consult `ARCHIVE_INDEX.md` or archive cards,163 do a bounded read, and summarize reusable findings in a recall note.16410. Promote any recalled high-value finding into the hot path instead of keeping165 it archive-only.16611. Refresh `CURRENT_STATE.md` and `ACTIVE_TRACKER.*` so a new agent can resume167 without replaying the whole ledger.16812. If tracker files exist, run169 `python <path-to-this-skill>/scripts/validate_tracker.py <tracker_dir>`170 after editing them. Resolve the script path relative to this `SKILL.md`.171172## Deliverables173174When using this skill, produce:175- A concise landscape summary176- The most informative next contrasts, ranked177- The control requirements for each contrast178- The predicted gradient or decision impact for each contrast179- Explicit hypothesis updates after new evidence180- A statement of what moved into the active set, what stayed as reference, and181 what was archived out of the default reading path182- When archive recall happened: why recall was triggered, what was read, what183 reusable asset was found, and whether it changed the plan184185## Quality bar186187- `runs.csv` should store facts and quality flags, not conclusions.188- `contrasts.csv` should make it obvious what changed, what stayed controlled, what was predicted, and what was learned.189- `hypotheses.md` should state what would change the belief, not just the belief itself.190- If a contrast is weakly controlled, mark it inconclusive instead of pretending it resolved the question.191- `CURRENT_STATE.md` should be readable in one short sitting.192- `ACTIVE_TRACKER.*` should contain only the experiments with nontrivial current193 decision gradient.194- Historical completeness belongs in the ledger and archive, not in the default195 agent entry point.196- Archive should be indexed by problem value, not only by date or snapshot name.197- Useful archived findings should be promotable back into the hot path with198 minimal rewriting.199200See [references/REFERENCE.md](references/REFERENCE.md) for the full rubric and [references/TEMPLATES.md](references/TEMPLATES.md) for starter files and examples.