| name | loopany-core-artifacts |
| description | Core artifact map and lifecycle rules for `signal` (deferred/unverified attention), `task` (committed work with an outcome), and `learning` (scoped belief from evidence). Briefly covers note, mission, brief, person, and skill-proposal routing. Use for choosing a core kind, writing signal/task bodies, promoting signal->task, and closing tasks with `## Outcome`. Usually read after [[proactive-capture/SKILL.md]] routes you here. |
core artifacts
signal = noticed but deferred or unverified.
task = committed work that must close with evidence.
learning = scoped belief derived from multiple artifacts.
[[reflect/SKILL.md]] reads these later. Prefer fewer, stronger artifacts: concrete
observations, explicit reasons, and honest outcomes.
Core kind map
| Kind | Use when | Detailed skill |
|---|
signal | Something concrete is noticed, but deferred or unverified | this file |
task | Someone is committing to work or closing work with an outcome | this file |
learning | A pattern across artifacts supports a scoped belief | [[reflect/SKILL.md]] |
skill-proposal | A learning implies a concrete skill behavior change | [[reflect/SKILL.md]] (user reviews + applies) |
mission | Long-running pursuit the agent should advance over time | kind file + [[monthly-review/SKILL.md]] |
brief | Point-in-time summary for the user | kind file + [[relations/SKILL.md]] for cites |
note | Free-form fallback; rationale, research, or one-off document | [[new-concept/SKILL.md]] |
person | Human entity with aliases/contact timeline | kind file |
This file gives full discipline for signal and task, plus a short guard
for learning. For new domain-specific entities, use
[[new-concept/SKILL.md]].
Decision rule
| State | Artifact |
|---|
| Acting now / committed to act | task |
| Deferred, blocked, or pending confirmation | signal |
| Pattern-backed belief from multiple artifacts | learning via [[reflect/SKILL.md]] |
| Belief implies a skill edit | skill-proposal via [[reflect/SKILL.md]] |
| Long-running standing intention | mission |
| Point-in-time summary | brief |
| Human entity | person |
| Pure observation with no unresolved action | nothing, or note if rationale matters |
| Belief without concrete evidence | wait for a pattern |
| Duplicate of an existing artifact | append evidence; don't fork |
The boundary that matters: acting now -> task. Deferred or
unverified -> signal.
Signal
Create a signal only when all gates pass:
- Concrete observable: file path, session ID, PostHog event, commit,
conversation, customer report.
- Actionable in principle: a plausible future task could address it.
- Not already captured: search first.
- Not being acted on now: if you are committing, create a
task.
loopany artifact list --kind signal --contains "<key phrase>"
Near-match existing signal? Append evidence to it. Recurrence must stay
visible in one place.
Fields
title is the searchable handle:
- 160 chars or less.
- Lead with the failure mode or symptom.
- Name the concrete thing: file path, tool, PR, event, integration.
title: "upload_image base64-encodes video -> >5MB -> PayloadTooLargeError"
title: "video uploads sometimes fail"
Body is optional. Add it when future context matters:
Observed <when/where>. <mechanism in 1-2 sentences>.
**Risk:** <cost|stability|data-loss|etc.>
**Fix direction (not yet a task):** <sentence>.
Keep signal bodies cite-worthy. They are not scratch pads.
Closing signals
Use addressed when a responsible artifact resolved it; usually after a
task reaches done with Outcome:
loopany artifact status <sig-id> addressed --addressed-by <artifact-id>
Use dismissed when closing without action:
- False positive after investigation.
- Duplicated by a stronger signal you are keeping.
- Underlying condition went away.
Always include a reason. Dismissed signals are still evidence.
For recurrences, append; don't fork. Promote to task when recurrence
reaches 3 times in 2 weeks, or sooner if risk becomes urgent.
Task
A task means "I'm doing this." Check for an existing task first:
loopany artifact list --kind task --contains "<key phrase>"
Near-duplicate in todo or running? Append or link with follows-up;
don't fork the thread.
Body
## Outcome
<What happened, final state, and any lesson or follow-up.>
## Outcome is required before done or failed. It should answer:
- What shipped, changed, or was decided?
- Is the issue fixed, or did the attempt fail?
- If there was an expected effect, did the observable evidence move?
- What would you do differently?
Optional sections are allowed when they make the Outcome easier to judge:
## Before for baseline evidence.## Plan or ## Fix for work notes.## Follow-up for loose threads that are not yet a next task or a
learning.
Failed or neutral results are still useful evidence. Say so directly.
Status
| Status | Meaning | Requirement |
|---|
todo | Not started | none |
running | Started | don't use as a parking lot |
in_review | PR open or waiting externally | use instead of long-running blocked running |
done | Delivered or shipped | ## Outcome required |
failed | Attempted, did not work, giving up | ## Outcome required |
cancelled | Decided not to do or abandoned | reason required; minimal Outcome if mid-flight |
Do not convert failed work into cancelled to keep the list clean.
Terminal gate:
loopany artifact append <id> --section Outcome --content "..."
loopany artifact status <id> <done|failed> --reason "<one line>"
check_at
Set check_at only when there is a concrete future question to answer.
It is read by loopany followups --due today.
Missing check_at is better than a date with no review question.
Learning
A learning is a belief with scope and evidence, not a log entry. Write
one only when a pattern is supported by multiple artifacts; one bad task is
not enough.
Use [[reflect/SKILL.md]] for full writing rules. Minimum discipline:
- Title is the belief itself, as a declarative sentence.
evidence cites the tasks, signals, or proposals that support it.- Body should cover
## Observation, ## Evidence, ## Scope, and
## Check-at.
check_at should usually be 1-3 months out, with a concrete revalidation
question.- When understanding changes, create a new learning with
supersedes; do
not rewrite the old one.
A learning may stop at "now we know." Create a skill-proposal only when
the belief implies a concrete skill edit.
Relations
Signal -> task
When a signal becomes work, create a task but leave the signal open.
After the task closes with Outcome:
loopany artifact status sig-... addressed --addressed-by tsk-...
This flips the signal and writes tsk-... addresses sig-....
Task -> task
Use follows-up when a new task continues a prior task:
loopany refs add --from tsk-future --to tsk-today --relation follows-up
Use follows-up for continuation. Use addresses for task -> signal.
See [[relations/SKILL.md]].
Task -> produced entity
Tasks are work-with-an-outcome. Tweets, posts, customers, orders, papers,
PRs, and episodes are entities that exist in the world. Represent them as
note or a custom kind via [[new-concept/SKILL.md]], then link from the
task with produced when needed.
Anti-patterns
Avoid:
- Just-in-case signals with weak evidence.
- Signals used as scratch pads.
- Vague titles like "announce lock issue".
- Dismissal without a reason.
- Duplicate signals instead of appended recurrence evidence.
- Same-day Outcomes for metrics that need time to settle.
- Outcomes that say only "Shipped. It works."
- Keeping externally blocked or abandoned work in
running.
- Using
task as an entity holder.
- One task -> one learning; wait for a pattern.
- Learning as a diary entry instead of a scoped belief.
- Learning without
check_at.
Command snippets
Create a signal:
loopany artifact create --kind signal \
--title "<one-line observable>" \
--domain <domain> \
--mentions "<entity-ids>" \
--content "$(cat <<'EOF'
Observed <when/where>. <mechanism in 1-2 sentences>.
**Risk:** <cost|stability|data-loss|etc.>
**Fix direction (not yet a task):** <sentence>.
EOF
)"
Append recurrence:
loopany artifact append sig-... --section "Recurrences" \
--content "2026-04-22: seen again in session X. Same mechanism."
Mark a signal addressed:
loopany artifact status <sig-id> addressed --addressed-by <tsk-id>
Dismiss a signal:
loopany artifact status <sig-id> dismissed --reason "..."
loopany artifact append <sig-id> --section "Dismissal" --content "..."
Create a scheduled task:
loopany artifact create --kind task \
--title "<...>" \
--status todo \
--check-at 2026-05-13
Close a task:
loopany artifact append <id> --section Outcome --content "..."
loopany artifact status <id> done --reason "<one line>"
