:sparkles: Tell agents to stage ephemeral files in /tmp by fabianvf · Pull Request #84 · konveyor/agentic-controller

fabianvf · 2026-07-31T16:35:15Z

Follow-up to #53, refs #70.

The agent's working directory is the git worktree the harness pushes at the end of a stage, but nothing tells the agent that, so it treats the repo as scratch space. Gave it a skill that only said "write the script to a file, make it executable, run it" — it wrote to /workspace/repo/verify.sh and left /tmp empty. Same skill with this change goes to /tmp.

Rule's in the harness rather than in each SkillCard since it's about the environment, not the skill. Emitted first so a skill that does say where to write can't override it.

Also moved prompt assembly out of main into internal/prompt while I was in there. Build is a pure function over a Layers struct and the prompt env vars moved into config.LoadFromEnv with the rest, so env reads stay in one package. Can split that out if you'd rather review the behaviour change on its own.

Summary by CodeRabbit

New Features
- Added layered prompt composition for agent, workflow, skill, and stage instructions.
- Added support for loading prompt context from environment configuration.
- Staging rules now direct ephemeral files to /tmp instead of the repository working tree.
- Workflow guidance supports a preferred setting with fallback compatibility.
- Added fallback guidance when no skill instructions are available.
Bug Fixes
- Prompts now consistently follow the intended layer order and newline formatting.
Tests
- Added coverage for prompt ordering, staging-rule precedence, empty layers, formatting, and configuration fallbacks.

coderabbitai · 2026-07-31T16:35:43Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 0f83551b-7ded-48dc-9516-fec5ec09600b

📥 Commits

Reviewing files that changed from the base of the PR and between f9ec140 and 261421c.

📒 Files selected for processing (7)

changes/unreleased/staging-rules.yaml
harness/cmd/migration-harness/main.go
harness/cmd/migration-harness/main_test.go
harness/internal/config/config.go
harness/internal/config/config_test.go
harness/internal/prompt/prompt.go
harness/internal/prompt/prompt_test.go

🚧 Files skipped from review as they are similar to previous changes (5)

changes/unreleased/staging-rules.yaml
harness/internal/config/config.go
harness/cmd/migration-harness/main.go
harness/internal/prompt/prompt_test.go
harness/internal/config/config_test.go

📝 Walkthrough

Walkthrough

The migration harness now uses internal/prompt to build ordered stage prompts from environment-backed context and stage data. The prompt includes staging rules and is submitted to ACP.

Changes

Layered prompt assembly

Layer / File(s)	Summary
Prompt context loading `harness/internal/config/config.go`, `harness/internal/config/config_test.go`	`Config` now loads agent, workflow, and stage instruction context. Workflow loading prefers `KONVEYOR_WORKFLOW_GUIDE` and falls back to `KONVEYOR_PLAYBOOK_INSTRUCTIONS`.
Prompt layers and composition `harness/internal/prompt/prompt.go`, `harness/internal/prompt/prompt_test.go`, `changes/unreleased/staging-rules.yaml`	The new package defines `Layers` and `Build`. `Build` emits staging rules first, omits empty layers, orders content from least to most specific, applies no-skill fallback guidance, and enforces one trailing newline.
Harness stage prompt submission `harness/cmd/migration-harness/main.go`	The migration harness builds `stagePrompt` through `prompt.Build`, submits it to ACP, and removes the local prompt builder.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant Config
  participant migrationHarness
  participant promptPackage
  participant ACP
  Config->>migrationHarness: provide environment-backed prompt context
  migrationHarness->>promptPackage: build stagePrompt from configured layers
  promptPackage-->>migrationHarness: return ordered stagePrompt
  migrationHarness->>ACP: submit stagePrompt

Possibly related PRs

konveyor/agentic-controller#83: Refactors skill-aware prompt construction into internal/prompt.Build while preserving no-skill fallback behavior.
konveyor/agentic-controller#36: Adds related playbook guide and stage instruction propagation.
konveyor/agentic-controller#80: Adds related workflow guide environment handling.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 28.57% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title uses the required ✨ alias and clearly states the primary change: staging ephemeral files in /tmp.
Description check	✅ Passed	The description explains the behavior change, design rationale, prompt refactor, and references the related issues; the required changelog fragment is present.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

Warning

There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure.

🔧 golangci-lint (2.12.2)

Error: build linters: plugin(logcheck): plugin "logcheck" not found
The command is terminated due to an error: build linters: plugin(logcheck): plugin "logcheck" not found

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

fabianvf · 2026-07-31T16:59:17Z

+Write ephemeral files to /tmp, never into the repository working tree:
+  - scripts you need to run: write them to /tmp, make them executable there,
+    and run them from there
+  - scratch notes, plans, logs, and intermediate output


This is the bit that makes scripts work. Rest of the prompt is unchanged, output's byte-identical to main once you strip this block off the front.

djzager

Nice refactor — the Layers struct and pure Build function are clean. Six comments, mostly about naming coordination with #80 and a couple of prompt-text suggestions.

The agent's working directory is the git worktree whose commits the harness pushes to the user's branch at the end of a stage. Nothing tells the agent that, so it treats the worktree as scratch space. Given a skill that said only "write the script to a file, make it executable, run it", claude-sonnet-5 wrote to /workspace/repo/verify.sh and left /tmp empty. Re-running the same skill with only this change: before: tool: write · /workspace/repo/verify.sh after: tool: shell · cat > /tmp/verify.sh << 'EOF' The rule goes in the harness rather than in each SkillCard because it describes the execution environment, not any particular skill — relying on every skill author to restate it is how it gets forgotten. It is emitted first so a skill that does say where to write cannot read as overriding it. Prompt assembly moves out of main into internal/prompt, alongside the other harness packages. Build is a pure function over an explicit Layers struct; the prompt env vars join the rest in config.LoadFromEnv, so env access stays in one package and the composition is testable without touching the environment. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Fabian von Feilitzsch <fabian@fabianism.us>

Rename the config fields to say what they are: AgentPrompt, WorkflowGuide, StageInstructions. cfg.Instructions was ambiguous next to the model and provider fields. Read KONVEYOR_WORKFLOW_GUIDE, falling back to KONVEYOR_PLAYBOOK_INSTRUCTIONS. that merge rather than depending on merge order. Drop the fallback once konveyor#80 has landed everywhere. Correct the staging rules: the harness commits .gitignore and .konveyor/analysis.json itself, so "you decide what gets committed" overstated the agent's control. Rename the ## Migration Context header to ## Workflow Guide. The prompt package is general purpose and shouldn't hardcode migration. Always end the prompt with exactly one newline. Sections appended "\n\n" but StageTask appended nothing, so the ending varied with whether a stage task was set. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Fabian von Feilitzsch <fabian@fabianism.us>

djzager

ACK

fabianvf force-pushed the staging-rules branch from ca6b7ae to 6f0ff73 Compare July 31, 2026 16:49

fabianvf commented Jul 31, 2026

View reviewed changes

djzager reviewed Jul 31, 2026

View reviewed changes

fabianvf and others added 2 commits August 3, 2026 10:57

fabianvf force-pushed the staging-rules branch from f9ec140 to 261421c Compare August 3, 2026 14:58

djzager approved these changes Aug 3, 2026

View reviewed changes

djzager merged commit 9b5bcad into konveyor:main Aug 3, 2026
15 checks passed

fabianvf deleted the staging-rules branch August 3, 2026 15:29

coderabbitai Bot mentioned this pull request Aug 4, 2026

✨ Add parameter injection to harness prompt #99

Closed

fabianvf mentioned this pull request Aug 10, 2026

Spike: executable skills — scripts and tooling in SkillCard/SkillCollection #70

Closed

coderabbitai Bot mentioned this pull request Aug 11, 2026

✨ Symlink skills dir for native goose discovery (#70) #134

Closed

coderabbitai Bot mentioned this pull request Aug 20, 2026

✨ Skill packaging and delivery, and rules that reach the model #157

Merged

Navigation Menu

✨ Tell agents to stage ephemeral files in /tmp - #84

✨ Tell agents to stage ephemeral files in /tmp#84
djzager merged 2 commits into
konveyor:mainfrom
fabianvf:staging-rules

fabianvf commented Jul 31, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Jul 31, 2026 •

edited

Loading

Walkthrough

Changes

Sequence Diagram(s)

❌ Failed checks (1 warning)

Uh oh!

fabianvf Jul 31, 2026

Uh oh!

djzager left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

djzager left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

fabianvf commented Jul 31, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Jul 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Sequence Diagram(s)

❌ Failed checks (1 warning)

Uh oh!

fabianvf Jul 31, 2026

Choose a reason for hiding this comment

Uh oh!

djzager left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

djzager left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

fabianvf commented Jul 31, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Jul 31, 2026 •

edited

Loading