@senior: De-duplicate agent model IDs (registry vs provider-config drift) #30
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: gitea/agents#30
Reference in New Issue
Block a user
Summary
Make the agent→model registry in
route.shthe single source of truth for Ollama Cloud model IDs, so the provider config ininstall-opencode.shcan no longer drift from it silently.route.shnow publishes the full registry as a step output;install-opencode.shderives theollama-cloudprovider'smodels:map from it (stripping theollama-cloud/prefix). No agent's model changed, and the routing/trigger-gate logic is untouched.Changes
route.sh: emitagents=$(jq -c '.' /tmp/agents.json)toGITHUB_OUTPUT(single-line, valid step-output value).agent.yml: passsteps.prep.outputs.agentsasAGENTSenv to the install-opencode step (mirrors the existingSKILLSpattern).install-opencode.sh: deriveCMODELSfrom$AGENTSand use it as theollama-cloudprovider'smodels:map;// {}keeps the JSON valid when no agent uses an ollama-cloud model (e.g. an all-anthropic caller repo). Removed the now-redundant hardcoded model list.Resolves #29 · 🤖 @senior
The agent→model registry in route.sh is now the single source of truth for Ollama Cloud model IDs. route.sh publishes the full registry as a step output (agents=...), and install-opencode.sh derives the ollama-cloud provider models: map from it (unprefixed IDs), instead of hardcoding a second copy. The two files can no longer drift silently — adding/changing a cloud model in the registry automatically updates the provider config. - route.sh: emit agents=$(jq -c '.' /tmp/agents.json) to GITHUB_OUTPUT - agent.yml: pass steps.prep.outputs.agents as AGENTS env to install-opencode.sh - install-opencode.sh: derive CMODELS from $AGENTS; // {} keeps it valid when no agent uses an ollama-cloud model (non-ollama repos)🤖 @senior — ✅ PR ready for review — @ffaerber please review & merge:
🤖 @senior — activity log (60 entries):
tool calls & reasoning
🔧 read:
.gitea/workflows/scripts/route.sh🔧 read:
.gitea/workflows/scripts/install-opencode.sh🔧 read:
.gitea/workflows/agent.yml🔧 bash:
cd /workspace/ffaerber/agents && git log --oneline -15 && echo "---BRANCH---" && git branch --show-current && echo "---STATUS---" && git status --short🔧 bash:
cd /workspace/ffaerber/agents && cat > /tmp/agents.json <<'JSON' { "pm": {"model":"ollama-cloud/gemma4:cloud","vision":true}, "junior": {"model":"ollama-cloud/kimi-k2.7-code:cloud","vision":false}, "senior": {"model":"ollama-cloud…💬 Let me verify my approach with the actual data before editing.
🔧 bash:
cd /workspace/ffaerber/agents echo "--- test1: build object ---" jq -c '[.[].model | select(startswith("ollama-cloud/")) | sub("^ollama-cloud/";"")] | unique | map({(.):{}}) | add' /tmp/agents.json echo "--- test2 empty ---" echo '{"x":{"mo…🔧 bash:
cd /workspace/ffaerber/agents echo "--- test empty with coalesce ---" echo '{"x":{"model":"anthropic/foo"}}' | jq -c '([.[].model | select(startswith("ollama-cloud/")) | sub("^ollama-cloud/";"")] | unique | map({(.):{}}) | add) // {}' echo …🔧 edit:
.gitea/workflows/scripts/route.sh💬 Verified.
// {}handles the empty case so non-ollama repos don't break. Now the edits — single source of truth: route.sh owns the registry, install-opencode.sh derives the cloud providermodels:map from it (via a step output, matching the existingSKILLSprecedent).🔧 edit:
.gitea/workflows/agent.yml🔧 edit:
.gitea/workflows/scripts/install-opencode.sh🔧 read:
.gitea/workflows/scripts/install-opencode.sh💬 That edit came out garbled with a leftover
$AGENTS_FILEreference. Let me fix it.🔧 edit:
.gitea/workflows/scripts/install-opencode.sh🔧 edit:
.gitea/workflows/scripts/install-opencode.sh🔧 edit:
.gitea/workflows/scripts/install-opencode.sh💬 Also simplify the verbose comment now that the code is one line:
🔧 read:
.gitea/workflows/scripts/install-opencode.sh🔧 read:
.gitea/workflows/scripts/route.sh💬 Now let me verify the final files and end-to-end that the generated config matches the old hardcoded one:
🔧 bash: `cd /workspace/ffaerber/agents
Simulate route.sh's agents.json output
AGENTS=$(jq -c '.' /tmp/agents.json)
echo "AGENTS=$AGENTS"
Run the derivation from install-opencode.sh
CMODELS=$(printf '%s' "$AGENTS" | jq -c '([.[].model | select(st…
💬 Now an end-to-end simulation comparing the generated provider config to the old hardcoded one: 🔧 **bash**:cd /workspace/ffaerber/agents && bash -n .gitea/workflows/scripts/route.sh && echo "route.sh: OK" && bash -n .gitea/workflows/scripts/install-opencode.sh && echo "install-opencode.sh: OK"echo "--- AGENTS line length check (GITHUB_OUTPUT mu…
💬 Derived models exactly match the old hardcoded set. Empty-case yields{}. Now check for shell syntax issues withbash -n: 💬 Both scripts parse cleanly; theagents=output line is single-line (319 chars, no newlines) so it's a valid GITHUB_OUTPUT value. Let me also confirm there's no other consumer of the old hardcoded model list, then commit and push. 🔧 **grep**:gemma4|kimi-k2|minimax-m3|glm-5.2🔧 **bash**:cd /workspace/ffaerber/agents && git add .gitea/workflows/scripts/route.sh .gitea/workflows/scripts/install-opencode.sh .gitea/workflows/agent.yml && git status --short && echo "---DIFF STAT---" && git diff --cached --stat💬 Only the registry in route.sh now names the cloud model IDs — exactly the single source of truth we wanted. Committing and pushing. 🔧 **bash**:cd /workspace/ffaerber/agents && git commit -m "@senior: issue #29 — derive Ollama Cloud provider models from the agent registryThe agent→model registry in route.sh is now the single source of truth for
Ollama Cloud model IDs. route.sh pu…
💬 Pushed toai/issue-29. The change makes route.sh's agent registry the single source of truth for Ollama Cloud model IDs: it now publishes the registry as a step output, and install-opencode.sh derives the providermodels:map from it (unprefixed IDs), so the two files can no longer drift. Verified the derived set exactly matches the previously-hardcoded models, and that the// {}` empty-case keeps the config valid for repos where no agent uses an ollama-cloud model.BEGIN_PR_DESCRIPTION
Summary
Make the agent→model registry in
route.shthe single source of truth for Ollama Cloud model IDs, so the provider config ininstall-opencode.shcan no longer drift from it silently.route.shnow publishes the full registry as a step output;install-opencode.shderives theollama-cloudprovider'smodels:map from it (stripping theollama-cloud/prefix). No agent's model changed, and the routing/trigger-gate logic is untouched.Changes
route.sh: emitagents=$(jq -c '.' /tmp/agents.json)toGITHUB_OUTPUT(single-line, valid step-output value).agent.yml: passsteps.prep.outputs.agentsasAGENTSenv to the install-opencode step (mirrors the existingSKILLSpattern).install-opencode.sh: deriveCMODELSfrom$AGENTSand use it as theollama-cloudprovider'smodels:map;// {}keeps the JSON valid when no agent uses an ollama-cloud model (e.g. an all-anthropic caller repo). Removed the now-redundant hardcoded model list.END_PR_DESCRIPTION
Pull request closed