Compare commits

...
Author SHA1 Message Date
issue-fix-bot ebb03e7063 fix: [alert] KubePodNotReady (paperless) (#896)
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
2026-09-21 09:45:31 +00:00
k.renner 63701fbebb Merge pull request 'ci(wiki-maintenance): verify the PR merged instead of trusting the merge status' (#895) from fix/wiki-maintenance-merge-verify into master
ci/woodpecker/manual/wiki-maintenance Pipeline was successful
ci/woodpecker/cron/e2e Pipeline failed
ci/woodpecker/cron/security-scan Pipeline was successful
ci/woodpecker/cron/issue-fix Pipeline was successful
ci/woodpecker/cron/alert-monitor Pipeline was successful
Reviewed-on: k.renner/homelab#895
2026-09-21 08:49:10 +00:00
k.renner e8a047eb14 Merge pull request 'ci(pick-model): newest among comparable candidates, policy in code' (#894) from feat/pick-model-newest-among-comparable into master
Reviewed-on: k.renner/homelab#894
2026-09-21 08:48:58 +00:00
claude-botandClaude Fable 5.1 0279210595 ci(wiki-maintenance): verify the PR merged instead of trusting the merge status
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
Run 10254: POST .../merge with delete_branch_after_merge returned
HTTP 500 although PR #2 on k.renner/notes was merged, so the step
failed and left the branch behind. The merge call no longer asks Gitea
to delete the branch; the step reads the PR back and requires
merged=true, then deletes the branch as a separate best-effort call.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 10:06:23 +02:00
claude-botandClaude Fable 5.1 25f09e9b52 ci(pick-model): newest among comparable candidates, policy in code
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
First live pick after #893 chose claude-sonnet-4-5 at 0.23 over
claude-sonnet-4-6: same-family models split the probability for a
mid-weight task, and "prefer newer" is a policy, not a judgment. New
scripts/ci/choose-model.sh applies it to the distribution: among
candidates within CHOOSE_MODEL_MARGIN (0.7) of the best probability the
most recently released wins; a clear best stays; NONE yields nothing.
The picker logs the top probabilities. Four policy tests added.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 09:59:40 +02:00
k.renner 82f2825d3f Merge pull request 'ci(pick-model): choose the model with one TypeSafe Choice, claude as fallback' (#893) from feat/pick-model-typesafe into master
ci/woodpecker/manual/wiki-maintenance Pipeline failed
ci/woodpecker/cron/issue-fix Pipeline was successful
ci/woodpecker/cron/alert-monitor Pipeline was successful
Reviewed-on: k.renner/homelab#893
2026-09-21 07:56:11 +00:00
claude-botandClaude Fable 5.1 0fec080b4d ci(pick-model): choose the model with one TypeSafe Choice, claude as fallback
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
The picker ran a full claude -p in plan mode per selection and parsed
JSON out of prose. With TYPESAFE_API_KEY set it now sends the task and
the catalog rows as state and asks one Choice over the model ids, each
with its catalog row as criterion plus NONE; the id is validated against
the pristine catalog as before. No key or a failed request falls through
to the claude picker unchanged.

Measured on the two-model fixture (Opus 5 / Haiku 4.5): issue-fix and
security-fix -> Opus 0.99, alert triage -> Opus 0.62, wiki audit ->
Haiku 0.94, update audit -> Haiku 0.46, distill -> Haiku 0.63; about a
second per pick. Dry-run test added to test-model-selection.sh.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 09:42:39 +02:00
k.renner 7424c6aefd Merge pull request 'ci(wiki-maintenance): rebase onto main before pushing, retry three times' (#892) from fix/wiki-maintenance-rebase-push into master
ci/woodpecker/manual/wiki-maintenance Pipeline was successful
ci/woodpecker/cron/issue-fix Pipeline was successful
ci/woodpecker/cron/alert-monitor Pipeline was successful
Reviewed-on: k.renner/homelab#892
2026-09-21 06:41:40 +00:00
claude-botandClaude Fable 5.1 4e3fb88309 ci(wiki-maintenance): push a branch and merge it through Gitea
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
Instead of rebasing onto a moving main: the run's changes go to
wiki-maintenance/<date>-<run>, a PR is opened and merged via the API
(Do: merge, branch deleted afterwards). A merge conflict leaves the PR
open for a human and fails the run. Fields checked against the
instance's swagger.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 08:40:47 +02:00
claude-botandClaude Fable 5.1 c8c5ce9015 ci(wiki-maintenance): rebase onto main before pushing, retry three times
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
Run 10241: distill and audit succeeded (5 sources, 8 new + 6 updated
notes), the push was rejected non-fast-forward because the workstation
pushed the vault while the agents ran. The push step now fetches,
rebases and retries; a real conflict still fails the run.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 08:36:58 +02:00
k.renner 293f51b914 Merge pull request 'ci(wiki-maintenance): strip CR/LF from secrets before use' (#891) from fix/wiki-maintenance-strip-secrets into master
ci/woodpecker/cron/alert-monitor Pipeline was successful
ci/woodpecker/manual/wiki-maintenance Pipeline failed
Reviewed-on: k.renner/homelab#891
2026-09-21 06:22:44 +00:00
claude-botandClaude Fable 5.1 70287db150 ci(wiki-maintenance): allow hand-started runs via the manual event
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
A deployment-event run (10239) errored before its first step: the
typesafe_api_key secret is scoped to cron and manual only. Manual
pipelines (woodpecker pipeline create) are the way to start this by
hand; the deployment trigger is dropped.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 08:20:36 +02:00
claude-botandClaude Fable 5.1 efedba1ff0 ci(wiki-maintenance): strip CR/LF from secrets before use
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was canceled
First cron run (10220) failed in checkout-notes with "could not read
Username": the wiki_bot_token secret had been set from a file with a
trailing newline, which git carried into the extraheader. The secret
is rewritten without it; the pipeline now strips CR/LF from the token
and the index credential regardless of how they were entered.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 08:19:12 +02:00
security-bot f35a2d5985 chore(image-automation): bump ghcr.io/paperless-ngx/paperless-ngx:3.2.1
ci/woodpecker/cron/wiki-maintenance Pipeline failed
ci/woodpecker/cron/e2e Pipeline failed
ci/woodpecker/cron/update-audit Pipeline was successful
ci/woodpecker/cron/issue-fix Pipeline was successful
ci/woodpecker/cron/alert-monitor Pipeline was successful
Updates managed by image-automation-controller.
2026-09-20 21:51:31 +00:00
k.renner 19b641a2b6 Merge pull request 'ci(wiki-maintenance): run on ci-claude 2.1.278' (#890) from fix/wiki-maintenance-image-2.1.278 into master
ci/woodpecker/cron/issue-fix Pipeline was successful
ci/woodpecker/cron/alert-monitor Pipeline was successful
Reviewed-on: k.renner/homelab#890
2026-09-20 20:46:19 +00:00
claude-botandClaude Fable 5.1 e422e6a13d ci(wiki-maintenance): run on ci-claude 2.1.278
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
The refresh triggered by #888 built 2.1.278 (the first ci-claude with
python3) rather than rebuilding 2.1.250, and its bump-pr step failed on
a newline in the push token secret (#889), so the consumer bump never
came. This pipeline needs python for wiki_ask.py, so it moves ahead of
the general bump.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-20 22:45:02 +02:00
k.renner f4d590c40d Merge pull request 'feat(ci): nightly wiki maintenance for the Notes vault as a Woodpecker cron' (#888) from feat/wiki-maintenance-ci into master
ci/woodpecker/push/build-ci-images Pipeline was successful
ci/woodpecker/push/ci-claude-refresh Pipeline failed
ci/woodpecker/cron/alert-monitor Pipeline was successful
Reviewed-on: k.renner/homelab#888
2026-09-20 20:41:58 +00:00
claude-botandClaude Fable 5.1 791cb6e524 feat(ci): nightly wiki maintenance for the Notes vault as a Woodpecker cron
ci/woodpecker/push/build-ci-images Pipeline was successful
ci/woodpecker/pr/ci-scripts-test Pipeline was successful
ci/woodpecker/pr/ci-manifests-test Pipeline was successful
ci/woodpecker/pr/pr-review Pipeline was successful
Replaces the workstation's wiki-maintenance.timer. The pipeline clones
k.renner/notes, runs wiki-distill over raw/ and wiki-audit over wiki/
as two separate `claude --agent` runs (no background tasks in print
mode, which killed the workstation job at the 600 s ceiling), then
commits and pushes as Claude Bot; the push webhook refreshes wiki-index.

The agent definitions now live in the vault (.claude/agents/), so CI
and workstation share them. The agents and their retrieval scripts
address the vault as ~/Notes; each agent step links $HOME/Notes to the
checkout. ci-claude gains python3 and py3-yaml for wiki_ask.py.

Secrets on the repo: claude_code_oauth_token (existing), typesafe_api_key,
wiki_bot_token (claude-bot, write:repository), wiki_index_basic_auth.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-20 22:37:59 +02:00
k.renner 3749df798e Merge pull request 'ci(wiki-index): per-commit image tags, deployment pinned to one' (#886) from fix/wiki-index-commit-tags into master
ci/woodpecker/push/build-ci-images Pipeline was successful
ci/woodpecker/cron/e2e Pipeline failed
ci/woodpecker/cron/security-scan Pipeline was successful
ci/woodpecker/cron/alert-monitor Pipeline was successful
ci/woodpecker/cron/issue-fix Pipeline was successful
Reviewed-on: k.renner/homelab#886
2026-09-20 09:00:25 +00:00
9 changed files with 317 additions and 3 deletions
+140
View File
@@ -0,0 +1,140 @@
# Nightly wiki maintenance for the Notes vault (k.renner/notes), replacing the
# workstation's wiki-maintenance.timer: distill the raw/ backlog into wiki/,
# then audit wiki/. Each agent is its own `claude` run, not a subagent of an
# orchestrator, so nothing depends on background tasks in print mode (the
# workstation job died on their 600 s ceiling). The pipeline commits and pushes
# as Claude Bot; the push webhook then refreshes the wiki-index service.
#
# The agents and their retrieval scripts address the vault as ~/Notes, so each
# agent step links $HOME/Notes to the checkout instead of rewriting them.
when:
- event: cron
cron: wiki-maintenance
# Hand-started runs: `woodpecker pipeline create k.renner/homelab --branch master`.
# The typesafe_api_key secret is scoped to cron and manual, not deployment.
- event: manual
steps:
- name: checkout-notes
image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
pull: true
environment:
WIKI_BOT_TOKEN:
from_secret: wiki_bot_token
commands:
- |
set -eu
# Token only on the wire, never in .git/config. Secrets set from files carry
# a trailing newline (run 10220: "could not read Username"); strip CR/LF.
TOKEN=$$(printf '%s' "$$WIKI_BOT_TOKEN" | tr -d '\r\n')
git -c http.extraheader="AUTHORIZATION: token $$TOKEN" \
clone --depth 50 --branch main https://gitea.renner.uno/k.renner/notes.git notes
echo "raw backlog: $$(ls notes/raw/*.md | grep -v _index.md | wc -l) file(s)"
- name: distill
image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
pull: true
environment:
CLAUDE_CODE_OAUTH_TOKEN:
from_secret: claude_code_oauth_token
TYPESAFE_API_KEY:
from_secret: typesafe_api_key
WIKI_INDEX_BASIC_AUTH:
from_secret: wiki_index_basic_auth
commands:
- |
set -eu
ln -sfn "$$PWD/notes" "$$HOME/Notes"
mkdir -p "$$HOME/.config/wiki-index"
printf '%s' "$$WIKI_INDEX_BASIC_AUTH" | tr -d '\r\n' > "$$HOME/.config/wiki-index/basic-auth"
chmod 600 "$$HOME/.config/wiki-index/basic-auth"
"$$HOME/Notes/bin/wiki_rank.py" --probe-index
- |
set -eu
MODEL=$$(sh scripts/ci/pick-model.sh "distill raw source notes into atomic wiki notes: read sources, check the wiki for existing coverage, write or merge notes; careful writing, moderate reasoning")
if [ -n "$$MODEL" ]; then echo "distill: selected model $$MODEL"; else echo "distill: using the CLI default model"; fi
cd notes
cat ../ci/agents/wiki-maintenance/distill.md | claude $${MODEL:+--model $$MODEL} \
--agent wiki-distill \
--permission-mode acceptEdits \
--allowedTools "Read,Edit,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*),Bash(mv /root/Notes/raw/:*),Bash(mkdir -p /root/Notes/:*)" \
--settings ../ci/agents/wiki-maintenance/settings.json \
-p
- name: audit
image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
pull: true
environment:
CLAUDE_CODE_OAUTH_TOKEN:
from_secret: claude_code_oauth_token
TYPESAFE_API_KEY:
from_secret: typesafe_api_key
WIKI_INDEX_BASIC_AUTH:
from_secret: wiki_index_basic_auth
commands:
- |
set -eu
ln -sfn "$$PWD/notes" "$$HOME/Notes"
mkdir -p "$$HOME/.config/wiki-index"
printf '%s' "$$WIKI_INDEX_BASIC_AUTH" | tr -d '\r\n' > "$$HOME/.config/wiki-index/basic-auth"
chmod 600 "$$HOME/.config/wiki-index/basic-auth"
- |
set -eu
MODEL=$$(sh scripts/ci/pick-model.sh "structural audit of ~1700 markdown notes: frontmatter completeness, broken links, orphans; mechanical checks, little reasoning")
if [ -n "$$MODEL" ]; then echo "audit: selected model $$MODEL"; else echo "audit: using the CLI default model"; fi
cd notes
cat ../ci/agents/wiki-maintenance/audit.md | claude $${MODEL:+--model $$MODEL} \
--agent wiki-audit \
--permission-mode acceptEdits \
--allowedTools "Read,Write,Glob,Grep,Bash(/root/Notes/bin/wiki_rank.py:*),Bash(/root/Notes/bin/wiki_ask.py:*)" \
--settings ../ci/agents/wiki-maintenance/settings.json \
-p
- name: push-notes
image: gitea.renner.uno/gitea_admin/ci-claude:2.1.278
pull: true
environment:
WIKI_BOT_TOKEN:
from_secret: wiki_bot_token
commands:
- |
set -eu
cd notes
git config user.email "claude-bot@renner.uno"
git config user.name "Claude Bot"
if git diff --quiet && git diff --cached --quiet && [ -z "$$(git ls-files --others --exclude-standard)" ]; then
echo "wiki-maintenance: no changes"
exit 0
fi
# The workstation pushes main on its own (notes-push.path), so main moves
# while the agents run (run 10241: non-fast-forward). The run's work goes
# to a branch and a PR that this step merges through Gitea; a conflict
# leaves the PR open for a human and fails the run.
BRANCH="wiki-maintenance/$$(date -u +%F)-$$CI_PIPELINE_NUMBER"
git checkout -q -b "$$BRANCH"
git add -A
git commit -q -m "wiki maintenance $$(date -u +%F): distill + audit (CI run $$CI_PIPELINE_NUMBER)"
TOKEN=$$(printf '%s' "$$WIKI_BOT_TOKEN" | tr -d '\r\n')
git -c http.extraheader="AUTHORIZATION: token $$TOKEN" push origin "HEAD:$$BRANCH"
API="https://gitea.renner.uno/api/v1/repos/k.renner/notes"
jq -n --arg title "wiki maintenance $$(date -u +%F) (CI run $$CI_PIPELINE_NUMBER)" \
--arg body "Nightly distill + audit. Merged by the pipeline; open only if the merge failed." \
--arg head "$$BRANCH" '{title: $$title, body: $$body, head: $$head, base: "main"}' > /tmp/pr.json
code=$$(curl -s -o /tmp/pr-resp.json -w '%{http_code}' -X POST \
-H "Authorization: token $$TOKEN" -H "Content-Type: application/json" --data @/tmp/pr.json "$$API/pulls")
echo "POST pull -> HTTP $$code"
[ "$$code" -ge 200 ] && [ "$$code" -lt 300 ] || { cat /tmp/pr-resp.json; exit 1; }
NUM=$$(jq -r '.number' /tmp/pr-resp.json)
# Merge without delete_branch_after_merge: run 10254 got HTTP 500 from that
# call although the merge itself had landed. Verify the PR state instead of
# trusting the status code, then drop the branch as a separate, best-effort step.
code=$$(curl -s -o /tmp/merge-resp.json -w '%{http_code}' -X POST \
-H "Authorization: token $$TOKEN" -H "Content-Type: application/json" \
--data '{"Do":"merge"}' "$$API/pulls/$$NUM/merge")
echo "POST merge #$$NUM -> HTTP $$code"
MERGED=$$(curl -s -H "Authorization: token $$TOKEN" "$$API/pulls/$$NUM" | jq -r '.merged')
echo "PR #$$NUM merged: $$MERGED"
[ "$$MERGED" = "true" ] || { cat /tmp/merge-resp.json; echo "PR #$$NUM left open"; exit 1; }
code=$$(curl -s -o /tmp/del-resp.json -w '%{http_code}' -X DELETE -H "Authorization: token $$TOKEN" \
"$$API/branches/$$(printf '%s' "$$BRANCH" | sed 's|/|%2F|g')")
echo "DELETE branch $$BRANCH -> HTTP $$code (best effort)"
+5
View File
@@ -0,0 +1,5 @@
You are running as the scheduled wiki-maintenance job in CI. The Notes vault is checked out at ~/Notes (the repository k.renner/notes).
Run the standard audit pass over ~/Notes/wiki/ and write the dated report to ~/Notes/output/. Do not modify any wiki note. Do not commit or push; the pipeline does that.
End with two lines: notes scanned, and issues found by category.
+7
View File
@@ -0,0 +1,7 @@
You are running as the scheduled wiki-maintenance job in CI. The Notes vault is checked out at ~/Notes (the repository k.renner/notes). Do exactly this, nothing else:
Glob ~/Notes/raw/*.md (top level only, never ~/Notes/raw/_processed/). Distill every file you find into ~/Notes/wiki/ following your process, including the ranker check for every concept. Move each fully processed source to ~/Notes/raw/_processed/. If raw/ has no top-level files besides _index.md, say so and stop.
Do not commit or push; the pipeline does that after you finish. Do not touch anything outside ~/Notes/wiki/ and ~/Notes/raw/.
End with one line per file: which wiki notes it produced or merged into, or "nothing distillable".
+37
View File
@@ -0,0 +1,37 @@
{
"//": "Permission scope for the wiki-maintenance agents (wiki-distill, wiki-audit) in CI.",
"//2": "The step's environment holds a Gitea push token, a TypeSafe key and the wiki-index",
"//3": "client credential, so the Read tool is kept away from /proc (env dumps), /root outside",
"//4": "the vault symlink target, /etc and the checkout's .git/. The vault itself is the work.",
"//5": "Mirrors ci/agents/issue-fix/settings.json; absolute globs because the vault is reached",
"//6": "through the ~/Notes symlink the pipeline creates.",
"permissions": {
"allow": [
"Read(//root/Notes/**)",
"Edit(//root/Notes/wiki/**)",
"Edit(//root/Notes/output/**)",
"Edit(//root/Notes/raw/**)",
"Bash(/root/Notes/bin/wiki_rank.py:*)",
"Bash(/root/Notes/bin/wiki_ask.py:*)",
"Bash(mv /root/Notes/raw/:*)",
"Bash(mkdir -p /root/Notes/:*)",
"Bash(ls:*)",
"Bash(date:*)",
"Bash(wc:*)",
"Bash(head:*)",
"Bash(tail:*)"
],
"deny": [
"Read(//proc/**)",
"Read(//etc/**)",
"Read(//root/.config/**)",
"Read(//root/Notes/.git/**)",
"Read(//root/Notes/_archive/**)",
"Edit(//root/Notes/_archive/**)",
"Bash(rm:*)",
"Bash(git:*)",
"Bash(curl:*)",
"Bash(wget:*)"
]
}
}
+3 -1
View File
@@ -13,7 +13,9 @@ RUN apk add --no-cache \
git \
jq \
openssh-client \
openssl
openssl \
python3 \
py3-yaml
# Install Claude Code, then make the resulting layer byte-stable so that a
# rebuild of the SAME version DEDUPS in the Gitea registry instead of orphaning
+28 -2
View File
@@ -27,6 +27,24 @@ data:
PAPERLESS_ALLOWED_HOSTS: "*"
PAPERLESS_TIME_ZONE: "Europe/Berlin"
PAPERLESS_OCR_LANGUAGE: "deu+eng"
# OCR concurrency, pinned rather than derived. paperless hands
# THREADS_PER_WORKER to ocrmypdf as its job count, and the default is
# cpu_count() / TASK_WORKERS. The container has no CPU limit, so
# multiprocessing.cpu_count() reports the NODE's cores — 20 on k8s-1 —
# and OCR fans out to "Starting processing with 20 workers concurrently".
# Twenty pages of a 300dpi colour scan rasterised at once blew past
# limits.memory below and the kernel OOMKilled the container (exit 137)
# mid-consume. Because celery re-picks the same document on every restart,
# that is a crash loop and not a one-off: the pod OOMs ~37s in, before the
# webserver ever finishes booting, so it never goes Ready and the archive
# stays down.
#
# TASK_WORKERS is already effectively 1 (20 cores / 1 worker = the 20 jobs
# observed); pinning it keeps that explicit so THREADS_PER_WORKER stays the
# only knob, and neither value silently changes when this lands on a node
# with a different core count.
PAPERLESS_TASK_WORKERS: "1"
PAPERLESS_THREADS_PER_WORKER: "4"
PAPERLESS_CONSUMER_RECURSIVE: "true"
PAPERLESS_CONSUMER_DELETE_DUPLICATES: "true"
PAPERLESS_CONSUMER_ENABLE_BARCODES: "true"
@@ -183,7 +201,7 @@ spec:
drop: ["ALL"]
containers:
- name: paperless
image: ghcr.io/paperless-ngx/paperless-ngx:3.2.0 # {"$imagepolicy": "flux-system:paperless-ngx"}
image: ghcr.io/paperless-ngx/paperless-ngx:3.2.1 # {"$imagepolicy": "flux-system:paperless-ngx"}
ports:
- name: http
containerPort: 8000
@@ -273,7 +291,15 @@ spec:
cpu: 200m
memory: 512Mi
limits:
memory: 2Gi
# Headroom, not the fix — capping OCR jobs in paperless-config
# is. Sized from what a healthy pod actually uses: the last
# long-lived container held a steady ~1.0Gi working set, so 2Gi
# left barely 1Gi for an OCR burst. 3Gi keeps a 4-job burst
# clear of the ceiling. The crashlooping containers never showed
# this in cadvisor (~820Mi peak) because they lived 37s against
# a ~30s scrape interval — the spike fell between samples, so
# the OOMKilled status, not the metric, is the reliable signal.
memory: 3Gi
securityContext:
runAsNonRoot: false
allowPrivilegeEscalation: true
+30
View File
@@ -0,0 +1,30 @@
#!/bin/sh
# Turn the TypeSafe picker's probability distribution into one model id.
#
# The judgment (how well each catalog model fits the task) comes from the
# model; the policy lives here, in code: among the candidates whose probability
# is within MARGIN of the best one, take the most recently released. Several
# same-family models (Sonnet 4.5 / 4.6 / 5) split the probability for a
# mid-weight task, and the newest of a comparable set is the one to run.
#
# Usage: choose-model.sh <catalog-table-file> < <systemone-response.json>
# The table is fetch-models.sh --table output: id \t name \t released=YYYY-MM-DD ...
# Prints the chosen id, or nothing when the answer is NONE / unusable.
# Env: CHOOSE_MODEL_MARGIN (default 0.7): candidate if p >= MARGIN * p_best.
set -eu
TABLE_FILE="${1:?usage: choose-model.sh <catalog-table-file> < response.json}"
MARGIN="${CHOOSE_MODEL_MARGIN:-0.7}"
jq -r --rawfile table "$TABLE_FILE" --argjson margin "$MARGIN" '
($table | split("\n") | map(select(length > 0) | split("\t")
| {key: .[0], value: ((.[2:] | map(select(startswith("released="))) | .[0] // "released=") | ltrimstr("released="))})
| from_entries) as $released
| (.answers.model.probabilities // {}) | to_entries
| map(select(.key != "NONE")) | sort_by(-.value)
| if length == 0 then empty
else (.[0].value) as $best
| map(select(.value >= $margin * $best))
| sort_by([($released[.key] // ""), .value])
| .[-1].key
end' 2>&1
+36
View File
@@ -33,6 +33,42 @@ none() { log "$1"; exit 0; } # emit nothing -> caller uses CLI defaul
TABLE="$(sh "$HERE/fetch-models.sh" --table || true)"
[ -n "$TABLE" ] || none "no catalog available — caller will use the CLI default model"
# First choice: one TypeSafe (System One) Choice over the catalog. The task and
# the catalog rows are state, every model id is an option with its catalog row
# as the criterion, and the answer is typed — no prompt, no JSON parsing, and a
# request that takes a fraction of a second instead of a claude run. The id is
# still validated against the pristine catalog. Falls through to the claude
# picker when the key is absent or the request fails.
if [ -n "${TYPESAFE_API_KEY:-}" ]; then
REQ="$(printf '%s\n' "$TABLE" | jq -R -s --arg task "$GUIDANCE" '
split("\n") | map(select(length > 0) | split("\t") | {id: .[0], name: .[1], info: (.[2:] | join(", "))})
| {model: "jev-latest",
state: {task: $task, models: .},
questions: {model: {
type: "choice",
instructions: "Which model in `models` should run `task`? Match the model to the task: not the largest or newest flagship for trivial or mechanical work, not a minimal model for hard, multi-step or security-sensitive work; prefer a newer model when otherwise comparable. Consider every listed model, including unfamiliar ones. Pick NONE only if no listed model fits.",
criteria: ((map({key: .id, value: "\(.name); \(.info)"}) | from_entries) + {NONE: "No listed model fits the task"})}}}')"
if [ -n "${PICK_MODEL_TYPESAFE_DRYRUN:-}" ]; then printf '%s\n' "$REQ"; exit 0; fi
RESP="$(printf '%s' "$REQ" | curl -sS --max-time 30 -X POST https://api.typesafe.ai/v1/systemone \
-H "Authorization: Bearer $TYPESAFE_API_KEY" -H 'Content-Type: application/json' --data-binary @- || true)"
# Policy in code (choose-model.sh): among candidates within a margin of the
# best probability, the most recently released wins. Same-family models split
# the probability for mid-weight tasks (live catalog: Sonnet 4.5 chosen at
# 0.23 over Sonnet 4.6).
TABLE_FILE="$(mktemp)"
printf '%s\n' "$TABLE" > "$TABLE_FILE"
CHOICE="$(printf '%s' "$RESP" | sh "$HERE/choose-model.sh" "$TABLE_FILE" | head -1)"
rm -f "$TABLE_FILE"
TOP="$(printf '%s' "$RESP" | jq -r '(.answers.model.probabilities // {}) | to_entries | sort_by(-.value) | .[0:4] | map("\(.key)=\(.value)") | join(" ")' 2>&1 | head -1)"
VALID="$(printf '%s\n' "$TABLE" | sh "$HERE/valid-model.sh" "$CHOICE")"
if [ -n "$VALID" ]; then
log "typesafe chose '$VALID' (top: ${TOP:-?}) for: $GUIDANCE"
printf '%s\n' "$VALID"
exit 0
fi
log "typesafe picker returned '${CHOICE:-<empty>}' (rejected, NONE, or request failed) — falling back to the claude picker"
fi
PROMPT="$(cat <<EOF
You are a model-selection agent for an automated CI job. From the models
currently available (listed below with their capabilities), choose the single
+31
View File
@@ -84,9 +84,40 @@ else
printf 'FAIL recorded fixture: opus=[%s] haiku=[%s]\n' "$opusline" "$haikuline"; fails=$((fails + 1))
fi
# TypeSafe path of pick-model.sh: the request offers exactly the catalog ids plus
# NONE as options, carries the task as state, and never sends anything else.
printf '%s' '{"data":[{"type":"model","id":"claude-opus-5","display_name":"Claude Opus 5","created_at":"2026-07-24T00:00:00Z","max_input_tokens":1000000,"max_tokens":128000},{"type":"model","id":"claude-haiku-4-5-20251001","display_name":"Claude Haiku 4.5","created_at":"2025-10-15T00:00:00Z","max_input_tokens":200000,"max_tokens":64000}],"has_more":false}' > "$FX"
req="$(FETCH_MODELS_BODY_FILE="$FX" TYPESAFE_API_KEY=dry PICK_MODEL_TYPESAFE_DRYRUN=1 sh "$HERE/pick-model.sh" "a task; with \"quotes\" and $ signs" 2>/dev/null || true)"
keys="$(printf '%s' "$req" | jq -r '.questions.model.criteria | keys | join(",")' 2>/dev/null || true)"
task="$(printf '%s' "$req" | jq -r '.state.task' 2>/dev/null || true)"
if [ "$keys" = "NONE,claude-haiku-4-5-20251001,claude-opus-5" ] && [ "$task" = 'a task; with "quotes" and $ signs' ]; then
printf 'ok typesafe request offers catalog ids + NONE and carries the task verbatim\n'
else
printf 'FAIL typesafe request: keys [%s] task [%s]\n' "$keys" "$task"; fails=$((fails + 1))
fi
# choose-model.sh: newest among comparable candidates, best otherwise, NONE -> nothing.
TB="$(mktemp)"
printf '%s\n' 'claude-sonnet-4-5 Claude Sonnet 4.5 released=2025-09-29 context=200000' \
'claude-sonnet-4-6 Claude Sonnet 4.6 released=2026-02-17 context=200000' \
'claude-opus-5 Claude Opus 5 released=2026-07-24 context=1000000' \
'claude-haiku-4-5 Claude Haiku 4.5 released=2025-10-15 context=200000' > "$TB"
choose() { printf '%s' "$1" | sh "$HERE/choose-model.sh" "$TB" || true; }
got="$(choose '{"answers":{"model":{"probabilities":{"claude-sonnet-4-5":0.30,"claude-sonnet-4-6":0.25,"claude-opus-5":0.10,"NONE":0.05}}}}')"
if [ "$got" = "claude-sonnet-4-6" ]; then printf 'ok newer sibling within margin wins\n'; else printf 'FAIL comparable: got [%s]\n' "$got"; fails=$((fails + 1)); fi
got="$(choose '{"answers":{"model":{"probabilities":{"claude-opus-5":0.90,"claude-haiku-4-5":0.05,"claude-sonnet-4-6":0.04,"NONE":0.01}}}}')"
if [ "$got" = "claude-opus-5" ]; then printf 'ok clear best is kept\n'; else printf 'FAIL clear best: got [%s]\n' "$got"; fails=$((fails + 1)); fi
got="$(choose '{"answers":{"model":{"probabilities":{"claude-haiku-4-5":0.60,"claude-sonnet-4-5":0.35,"NONE":0.05}}}}')"
if [ "$got" = "claude-haiku-4-5" ]; then printf 'ok older best outside margin is kept\n'; else printf 'FAIL margin: got [%s]\n' "$got"; fails=$((fails + 1)); fi
got="$(choose '{"answers":{"model":{"probabilities":{"NONE":1.0}}}}')"
if [ -z "$got" ]; then printf 'ok NONE yields nothing\n'; else printf 'FAIL NONE: got [%s]\n' "$got"; fails=$((fails + 1)); fi
rm -f "$TB"
if [ "$fails" -eq 0 ]; then
echo "All model-selection tests passed."
else
echo "$fails test(s) FAILED."
exit 1
fi