incident-capture
Skill0 starsChanged 5 months agoMITView on GitHub
When a component fires fallback_activated (or crashes) repeatedly in a window, auto-open a pre-filled incident note so the 2am page starts with context instead of a blank page.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- Nashova/LLM-Dev-Frameworktheir own version, first seen Nov 1, 2026
- Nitmi/embedded-agent-toolkittheir own version, first seen Sep 9, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- Nitmi/embedded-agent-toolkit★ 2same name
- Nashova/LLM-Dev-Framework★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository29 other files it gives its agents.
- adopt-existing-projectSkill
- adrSkill
- branch-conflict-checkSkill
- ci-watchSkill
- compound-learningsSkill
- config-auditSkill
- cost-checkSkill
- coverage-ratchetSkill
- doc-syncSkill
- eval-harnessSkill
- flaky-triageSkill
- followup-trackingSkill
- kit-doctorSkill
- kit-updateSkill
- logging-initSkill
- migration-checkSkill
- new-project-bootstrapSkill
- nightly-auditSkill
- observability-checkSkill
- parallel-workSkill
- perf-budgetSkill
- pr-babysitterSkill
- pre-prSkill
- prompt-regressionSkill
- releaseSkill
- session-handoffSkill
- sync-healthSkill
- using-the-kitSkill
- writing-kit-skillsSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 47
If the count is below the threshold, exit 0 — nothing to open.
- Distinctive: line 50
Before writing a new note, check whether an incident for this component is already active.
- Distinctive: line 65
Keep this section factual — log facts and timeline, not speculative root-cause filler.
- Distinctive: line 92
If the component last changed within the same git session, it is the primary suspect.
- Very specific: line 115
Commit on main (ops notes) or a short-lived branch — whichever your project allows.
- Distinctive: line 132
If the post fails, log the failure and continue — the note on disk is the source of truth.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Incident Capture
Profile-driven. Alert destination is alerts.channel / alerts.target in .claude/kit.yaml. The SQLite fallback queries below use the table named by capabilities.llm.spend_table (default: invocation_log); substitute if your project differs.
A single fallback_activated event is expected; three in fifteen minutes is an
incident. This skill detects the repetition pattern, gathers the correlated log
context and the recent commits touching the component, writes an incident note
from templates/incident_note.template.md, and posts a summary to chat. The
person who picks up the page gets a filled-in timeline instead of a blank
document. See the observability-check skill for the logging conventions this
depends on, and templates/fallback_alert.py for the fallback_activated event
contract and the FallbackNotifier dedup mechanism.
Workflow
1. Detect the trigger
Define "repeated" as N occurrences of the same event_type for the same
component within a rolling window. Sensible defaults: N = 3, window = 15 m.
Query Loki, the local structured log table, or the SQLite invocation log —
whichever is available:
# Loki (LogQL) — adapt label selectors to your stack
logcli query \
'{app="donna"} |= "fallback_activated" | json | component="<component>"' \
--since=15m --output=jsonl \
| jq -s 'length'
# SQLite fallback — table: capabilities.llm.spend_table (kit.yaml)
sqlite3 donna_tasks.db "
SELECT COUNT(*) FROM invocation_log
WHERE event_type = 'fallback_activated'
AND component = '<component>'
AND created_at >= datetime('now', '-15 minutes');"If the count is below the threshold, exit 0 — nothing to open.
2. Deduplicate — one open incident per component
Before writing a new note, check whether an incident for this component is already active. A sentinel file naming convention keeps the check simple:
mkdir -p docs/incidents/
ls docs/incidents/ 2>/dev/null | grep -q "^${component}-.*-open\.md$" && exit 0This prevents a flapping component from generating a new note on every poll
cycle. When the existing incident is resolved, rename the file (swap -open.md
for -resolved.md) to re-arm the trigger.
3. Gather correlated log context
Pull the log lines for the window — include correlation_id, the error, and
the fallback taken. These become the "Correlated logs" section of the note.
Keep this section factual — log facts and timeline, not speculative
root-cause filler. The commands below already cap the excerpt at 20 lines
from the 15-minute window:
logcli query \
'{app="donna"} |= "fallback_activated" | json | component="<component>"' \
--since=15m --output=jsonl \
| jq -r '"\(.ts) [\(.correlation_id // "no-id")] \(.error) → fallback: \(.fallback)"' \
| head -20
# SQLite fallback — table: capabilities.llm.spend_table (kit.yaml)
sqlite3 -csv donna_tasks.db "
SELECT created_at, correlation_id, error, fallback
FROM invocation_log
WHERE event_type = 'fallback_activated'
AND component = '<component>'
AND created_at >= datetime('now', '-15 minutes')
ORDER BY created_at;" | head -204. Find the last commits touching the component
# Replace the path with the actual module and config locations
git log --oneline -10 -- "src/<pkg>/<component>/" "config/<component>*.yaml"If the component last changed within the same git session, it is the primary suspect. These hashes go into the "Related commits" section of the note. This skill documents the incident; it does not revert or patch the primary-suspect commit — that decision belongs to whoever responds.
5. Write the incident note from the template
component="<component>"
slug="${component}-$(date -u +%Y%m%d-%H%M%S)-open"
note="docs/incidents/${slug}.md"
# Populate with envsubst, sed, or a project helper; the template uses
# <PLACEHOLDER> markers — see templates/incident_note.template.md.
INCIDENT_COMPONENT="$component" \
INCIDENT_COUNT="$count" \
INCIDENT_WINDOW="15m" \
INCIDENT_TIMESTAMP="$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
envsubst < templates/incident_note.template.md > "$note"
git add "$note"
git commit -m "incident: open ${component} $(date -u +%Y-%m-%d)"Commit on main (ops notes) or a short-lived branch — whichever your project
allows. Log a follow-up ID in docs/followups.md via the followup-tracking
skill if the incident uncovers deferred work.
6. Post to chat
Send a brief alert to the debug/ops channel with a direct link to the note:
# kit.yaml → alerts.channel / alerts.target (destination; swap URL/payload for Slack)
curl -sX POST "$DISCORD_WEBHOOK_URL" \
-H 'Content-Type: application/json' \
-d "$(jq -n \
--arg msg "**Incident opened** · \`${component}\` fired \`fallback_activated\` ${count}× in ${window}
→ \`${note}\`" \
'{content: $msg}')"If the post fails, log the failure and continue — the note on disk is the source of truth. Never let a broken notifier prevent the note from being written.
Output
## Incident Capture
- Component: <name>
- Trigger: <event_type> × <count> in <window>
- Note: docs/incidents/<slug>-open.md
- Correlation IDs: <list>
- Related commits: <short hashes + subjects | none in window>
- Chat notified: yes | failed (<reason>)
- Deduped: no — new note created | yes — existing open incident, skippedDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

