Conversation
Contributor
🏷️ Automatic Labeling SummaryThis PR has been automatically labeled based on the files changed and PR metadata. Applied Labels: size-xs Label Categories
For more information, see |
Contributor
🔍 Lighthouse Performance Audit
📥 Download full Lighthouse report Budget Compliance: Performance budgets enforced via |
…tputs session idle timeout Agent-Logs-Url: https://github.com/Hack23/riksdagsmonitor/sessions/f6b0a915-3add-46af-85c2-dd74cd7f0051 Co-authored-by: pethers <1726836+pethers@users.noreply.github.com>
github-actions
Bot
added
documentation
Contributor
🔍 Lighthouse Performance Audit
📥 Download full Lighthouse report Budget Compliance: Performance budgets enforced via |
…udget Agent-Logs-Url: https://github.com/Hack23/riksdagsmonitor/sessions/f6b0a915-3add-46af-85c2-dd74cd7f0051 Co-authored-by: pethers <1726836+pethers@users.noreply.github.com>
Copilot
AI
changed the title
[WIP] Fix workflow failure for government propositions
fix(aw): front-load propositions PR call before safeoutputs idle timeout
Apr 23, 2026
Contributor
🔍 Lighthouse Performance Audit
📥 Download full Lighthouse report Budget Compliance: Performance budgets enforced via |
Contributor
🔍 Lighthouse Performance Audit
📥 Download full Lighthouse report Budget Compliance: Performance budgets enforced via |
Contributor
There was a problem hiding this comment.
Pull request overview
Updates the News: Government Propositions agentic workflow prompt to avoid Safe Outputs “session not found” failures by moving the safeoutputs___create_pull_request intent-recording call earlier in the run, aligning the time budget with the observed Streamable-HTTP idle timeout.
Changes:
- Added a prominent safeoutputs idle-timeout callout and clarified that the PR intent must be recorded early.
- Compressed Run 1 (analysis) and Run 2 (articles) time budgets so the PR call happens well before the observed cutoff.
- Added an early-exit rule to stage/commit and open an
[early-pr]PR if staging hasn’t started by minute 25.
| **Run 1 — Analysis mode** (no prior analysis found, ~50 min — produces all 23 artifacts): | ||
| > 🔴 **CRITICAL — safeoutputs MCP idle timeout (~30 min)**: The `safeoutputs` MCP server runs a Streamable-HTTP session that expires after **~30–35 minutes of idle time** (confirmed in run [24820030825](https://github.com/Hack23/riksdagsmonitor/actions/runs/24820030825) where all `create_pull_request` calls at minute ~32 failed with `session not found`, losing 37 files of completed analysis work). `sandbox.mcp.keepalive-interval: 300` keeps TCP connections alive but does **NOT** refresh the server-side session. **Your first and only `safeoutputs___*` call MUST happen before minute 28 of agent time.** This is a harder deadline than the ~60-minute Copilot-API token window described in `00-base-contract.md §Session keepalive requirement` and overrides the generic 48-min deadline in `07-commit-and-pr.md §Deadline enforcement` for this workflow. | ||
| > | ||
| > **AI-FIRST within the compressed budget**: Pass 2 is still mandatory (AI FIRST principle, §"5. 🔴 AI FIRST Quality Principle" in repo custom instructions). Under the tightened ~28-min Run 1 budget, prefer **scope compression over iteration skipping** — reduce the number of per-document analyses if needed, but always perform a full read-back-and-improve Pass 2 on whatever artifacts exist. For scheduled runs treat `analysis_depth` as `standard` in practice even if the input defaults to `deep`; reserve `deep`/`comprehensive` for manual `workflow_dispatch` backfills where a human can monitor for a longer-session environment. A partial analysis with genuine iteration is always strictly better than a broader analysis that never reaches GitHub. |
There was a problem hiding this comment.
The guidance to “reduce the number of per-document analyses if needed” conflicts with the Analysis Gate’s per-document coverage requirement (one document analysis per dok_id in data-download-manifest.md, see 05-analysis-gate.md check 2). As written, following this advice will predictably fail the gate. Consider reframing scope compression as reducing the download/manifest scope (e.g., fewer dok_id entries) while still maintaining 1:1 per-document coverage for whatever remains in the manifest.
Suggested change
| > **AI-FIRST within the compressed budget**: Pass 2 is still mandatory (AI FIRST principle, §"5. 🔴 AI FIRST Quality Principle" in repo custom instructions). Under the tightened ~28-min Run 1 budget, prefer **scope compression over iteration skipping** — reduce the number of per-document analyses if needed, but always perform a full read-back-and-improve Pass 2 on whatever artifacts exist. For scheduled runs treat `analysis_depth` as `standard` in practice even if the input defaults to `deep`; reserve `deep`/`comprehensive` for manual `workflow_dispatch` backfills where a human can monitor for a longer-session environment. A partial analysis with genuine iteration is always strictly better than a broader analysis that never reaches GitHub. | |
| > **AI-FIRST within the compressed budget**: Pass 2 is still mandatory (AI FIRST principle, §"5. 🔴 AI FIRST Quality Principle" in repo custom instructions). Under the tightened ~28-min Run 1 budget, prefer **scope compression over iteration skipping** — reduce the **download/manifest scope** if needed (for example, fewer `dok_id` entries in `data-download-manifest.md`), but maintain **1:1 per-document coverage** for every `dok_id` that remains in the manifest and always perform a full read-back-and-improve Pass 2 on whatever artifacts exist. For scheduled runs treat `analysis_depth` as `standard` in practice even if the input defaults to `deep`; reserve `deep`/`comprehensive` for manual `workflow_dispatch` backfills where a human can monitor for a longer-session environment. A smaller manifest with complete per-document coverage and genuine iteration is always strictly better than a broader manifest that cannot clear the Analysis Gate or never reaches GitHub. |
| | 2–6 | Read all 23 analysis artifacts into context (Families A+B+C+D) | 06 | | ||
| | 6–18 | Article Pass 1 + Pass 2 (EN, SV) | 06 | | ||
| | 18–22 | Visual + link validation | 06 | | ||
| | 22–26 | Stage articles, commit, **ONE** `safeoutputs___create_pull_request` — **HARD DEADLINE minute 28** | 07 | |
There was a problem hiding this comment.
Run 2’s staging/PR row is 22–26 but labels a “HARD DEADLINE minute 28”. If minute 28 is the real cutoff (per the safeoutputs session constraint), consider extending the row to 22–28 or changing the hard-deadline label to minute 26 so the table is internally consistent.
Suggested change
| | 22–26 | Stage articles, commit, **ONE** `safeoutputs___create_pull_request` — **HARD DEADLINE minute 28** | 07 | | |
| | 22–28 | Stage articles, commit, **ONE** `safeoutputs___create_pull_request` — **HARD DEADLINE minute 28** | 07 | |
| | 22–26 | Stage articles, commit, **ONE** `safeoutputs___create_pull_request` — **HARD DEADLINE minute 28** | 07 | | ||
|
|
||
| Trim scope before quality. Never open a second PR within a run — there is no second PR. | ||
| Trim scope before quality. Never open a second PR within a run — there is no second PR. **If you reach minute 25 without staging, stop all remaining analysis/article work, commit whatever exists on disk (include `[early-pr]` in the commit message), and call `create_pull_request` immediately** — a partial-but-delivered PR is infinitely better than losing all work to a `session not found` error. |
There was a problem hiding this comment.
This paragraph instructs to call create_pull_request immediately, but elsewhere the workflow is explicit about safeoutputs___create_pull_request. Using the unqualified name risks the agent calling a non-existent/incorrect tool name. Consider changing this to safeoutputs___create_pull_request for consistency and to prevent tool-selection mistakes under time pressure.
Suggested change
| Trim scope before quality. Never open a second PR within a run — there is no second PR. **If you reach minute 25 without staging, stop all remaining analysis/article work, commit whatever exists on disk (include `[early-pr]` in the commit message), and call `create_pull_request` immediately** — a partial-but-delivered PR is infinitely better than losing all work to a `session not found` error. | |
| Trim scope before quality. Never open a second PR within a run — there is no second PR. **If you reach minute 25 without staging, stop all remaining analysis/article work, commit whatever exists on disk (include `[early-pr]` in the commit message), and call `safeoutputs___create_pull_request` immediately** — a partial-but-delivered PR is infinitely better than losing all work to a `session not found` error. |
| ## Time budget | ||
|
|
||
| **Run 1 — Analysis mode** (no prior analysis found, ~50 min — produces all 23 artifacts): | ||
| > 🔴 **CRITICAL — safeoutputs MCP idle timeout (~30 min)**: The `safeoutputs` MCP server runs a Streamable-HTTP session that expires after **~30–35 minutes of idle time** (confirmed in run [24820030825](https://github.com/Hack23/riksdagsmonitor/actions/runs/24820030825) where all `create_pull_request` calls at minute ~32 failed with `session not found`, losing 37 files of completed analysis work). `sandbox.mcp.keepalive-interval: 300` keeps TCP connections alive but does **NOT** refresh the server-side session. **Your first and only `safeoutputs___*` call MUST happen before minute 28 of agent time.** This is a harder deadline than the ~60-minute Copilot-API token window described in `00-base-contract.md §Session keepalive requirement` and overrides the generic 48-min deadline in `07-commit-and-pr.md §Deadline enforcement` for this workflow. |
There was a problem hiding this comment.
The callout says the first/only safeoutputs___* call MUST happen before minute 28, but the Run 1 schedule allocates 25–28 for staging + PR, which can imply calling at minute 28. To avoid conflicting guidance, either tighten the window (e.g. end at 27) or reword the callout to “by minute 28 at the latest / no later than minute 28”.
Suggested change
| > 🔴 **CRITICAL — safeoutputs MCP idle timeout (~30 min)**: The `safeoutputs` MCP server runs a Streamable-HTTP session that expires after **~30–35 minutes of idle time** (confirmed in run [24820030825](https://github.com/Hack23/riksdagsmonitor/actions/runs/24820030825) where all `create_pull_request` calls at minute ~32 failed with `session not found`, losing 37 files of completed analysis work). `sandbox.mcp.keepalive-interval: 300` keeps TCP connections alive but does **NOT** refresh the server-side session. **Your first and only `safeoutputs___*` call MUST happen before minute 28 of agent time.** This is a harder deadline than the ~60-minute Copilot-API token window described in `00-base-contract.md §Session keepalive requirement` and overrides the generic 48-min deadline in `07-commit-and-pr.md §Deadline enforcement` for this workflow. | |
| > 🔴 **CRITICAL — safeoutputs MCP idle timeout (~30 min)**: The `safeoutputs` MCP server runs a Streamable-HTTP session that expires after **~30–35 minutes of idle time** (confirmed in run [24820030825](https://github.com/Hack23/riksdagsmonitor/actions/runs/24820030825) where all `create_pull_request` calls at minute ~32 failed with `session not found`, losing 37 files of completed analysis work). `sandbox.mcp.keepalive-interval: 300` keeps TCP connections alive but does **NOT** refresh the server-side session. **Your first and only `safeoutputs___*` call MUST happen by minute 28 at the latest.** This is a harder deadline than the ~60-minute Copilot-API token window described in `00-base-contract.md §Session keepalive requirement` and overrides the generic 48-min deadline in `07-commit-and-pr.md §Deadline enforcement` for this workflow. |
Member
|
@copilot apply changes based on the comments in this thread |
Agent-Logs-Url: https://github.com/Hack23/riksdagsmonitor/sessions/1637228f-c40b-4690-a6ea-62aaa9b990fe Co-authored-by: pethers <1726836+pethers@users.noreply.github.com>
Contributor
Author
Applied all four reviewer suggestions in 5e49385:
|
Contributor
🔍 Lighthouse Performance Audit
📥 Download full Lighthouse report Budget Compliance: Performance budgets enforced via |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
News: Government Propositions(run 24820030825) produced no safe outputs: the agent completed 37 analysis files and committed locally at min 31, then allsafeoutputs___create_pull_requestcalls at min 32 failed withStreamable HTTP error: session not found.Root cause
The safeoutputs MCP uses a Streamable-HTTP session with a ~30–35 min idle timeout (prior PR #1895).
sandbox.mcp.keepalive-interval: 300keeps the TCP connection alive but does not refresh server-side session state. The workflow's Time budget directed the single PR call to min 47–50 — always past the cutoff.Changes (
.github/workflows/news-propositions.mdbody only)safeoutputs___*call MUST happen by minute 28 at the latest, and explicitly overriding the 48-min deadline in07-commit-and-pr.md §Deadline enforcementfor this workflow.[early-pr]tag and call the fully-qualifiedsafeoutputs___create_pull_requestimmediately.dok_identries indata-download-manifest.md) over iteration skipping, while maintaining 1:1 per-document coverage for everydok_idthat remains in the manifest so05-analysis-gate.mdcheck 2 still passes; Pass 2 remains mandatory; scheduled runs behave asanalysis_depth: standardin practice; reservedeep/comprehensivefor manualworkflow_dispatchbackfills.Notes for reviewers
{{#runtime-import .github/workflows/news-propositions.md}}; the lock tracksfrontmatter_hashonly.07-commit-and-pr.md §Core ruleforbids heartbeat/keep-alive PRs post-fix: call safeoutputs___create_pull_request ASAP, never noop when work exists #1768 (create-pull-request.max: 1). This aligns the per-workflow budget with00-base-contract.md §Session keepalive requirement("well before the session approaches expiry") rather than reviving the fix(aw): prevent safeoutputs session timeout in news-committee-reports #1895 pattern.tests/workflow-architecture.test.ts(80 tests) passes; no tests assert on time-budget text.