Merge pull request #2847 from Hack23/copilot/fix-agent-job-failure

refactor: remove repo-memory usage and lock-file patching, favor cache-memory

Author: pethers

.github/copilot-instructions.md

@@ -165,8 +165,9 @@ tools:
   agentic-workflows: true # Workflow introspection (status, compile, logs, audit, checks)
   bash: true              # Shell commands
   playwright:             # Browser automation (where needed)
-  repo-memory:            # Persistent memory across runs
-    branch-name: memory/news-generation
+  cache-memory:           # Session state persisted via GitHub Actions cache (retention configured per workflow, eviction-dependent)
+    key: news-${{ github.workflow }}-${{ inputs.article_date || 'today' }}
+    retention-days: 14
 ```
 
 ### MCP Server Configuration (All Workflows)

.github/prompts/00-base-contract.md

@@ -97,7 +97,7 @@ Print `agent_minute` (helper in [`01-bash-and-shell-safety.md §Time-budget self
 | Aggregate + render all 14 languages | **36 – 40** | Blocked at minute 40 → commit analysis-only and PR anyway. |
 | Stage + commit + sandbox handoff + PR | **40 – 42** | Hard deadline **45**. |
 
-No per-phase checkpoint PRs. No repo-memory push steps.
+No per-phase checkpoint PRs. No unnecessary memory push steps.
 
 ## Output contract
 

.github/prompts/07-commit-and-pr.md

@@ -149,6 +149,10 @@ Every news workflow declares `tools.cache-memory:` keyed by `news-${{ github.wor
 2. After a successful Pass 1 (or after the analysis gate passes), copy the produced `.md` artifacts back to `/tmp/gh-aw/cache-memory/$ARTICLE_DATE/$SUBFOLDER/` so they are available for persistence if the workflow later fails during PR publication or another post-agent stage.
 3. The agent only writes to `/tmp/gh-aw/cache-memory/`; no safe-output call persists cache-memory. The compiled workflow's cache-update step runs only after a successful agent job (`needs.agent.result == 'success'`), so recovery is reliable for **post-agent failures** (PR-publication problems) but **not** for agent-job failures or timeouts.
 
+**Checkout settings:** `fetch-depth: 1` is sufficient for `cache-memory`. gh-aw manages a self-contained git repo inside `/tmp/gh-aw/cache-memory/` independently of the main workspace. No special `checkout.fetch:` or `fetch-depth: 0` setting is required. Git user identity is configured by the dedicated "Configure Git credentials" step in compiled workflows, not by `actions/checkout`.
+
+**Resilience:** The compiled `Commit cache-memory changes` and `Check cache-memory git integrity` steps run with `if: always()`. If a corrupted or missing `.git` directory from a restored cache causes a transient failure, the subsequent integrity-check step detects and reseeds the git repo so the artifact upload can still succeed. Cache-memory failures should not block article generation — the primary output is always committed to `analysis/daily/`.
+
 Cache-memory is a recovery mechanism for the next run, not a substitute for committing real files on disk under `analysis/daily/`.
 
 ## PR creation resilience (`fallback-as-issue`, `if-no-changes`, host-side PAT fallback)

.github/skills/gh-aw-README.md

@@ -229,12 +229,9 @@ tools:
   agentic-workflows: true    # Workflow introspection (status, compile, logs, audit, checks)
   bash: true                 # Shell command execution
   playwright:                # Browser automation (optional, for visual validation)
-  repo-memory:               # Persistent memory across workflow runs
-    branch-name: memory/news-generation
-    allowed-extensions: [".md", ".json"]
-    max-file-size: 51200
-    max-file-count: 50
-    max-patch-size: 51200
+  cache-memory:              # Session state persisted via GitHub Actions cache (retention configured per workflow)
+    key: news-${{ github.workflow }}-${{ inputs.article_date || 'today' }}
+    retention-days: 14
 ```
 
 ### Available `agentic-workflows:` Tools

.github/skills/gh-aw-mcp-configuration/SKILL.md

@@ -1792,7 +1792,7 @@ For Copilot coding agent sessions (not agentic workflows), MCP servers are confi
 - [ ] Version MCP server APIs
 - [ ] Implement graceful degradation
 - [ ] Use timeout for all MCP operations
-- [ ] **Prefer `repo-memory:` and `cache-memory:` over `@modelcontextprotocol/server-memory`** — gh-aw native tools are persistent across runs; generic MCP memory is ephemeral per process
+- [ ] **Prefer `cache-memory:` over `@modelcontextprotocol/server-memory`** — gh-aw native cache-memory is persistent across runs (~7-14 days); generic MCP memory is ephemeral per process
 - [ ] **Skip `@modelcontextprotocol/server-sequential-thinking`** — modern LLMs (Claude Opus 4.8, GPT-5) have native CoT; it wastes context tokens
 
 ---

.github/skills/gh-aw-workflow-authoring/SKILL.md

@@ -1002,22 +1002,19 @@ Stagger schedules to avoid resource contention. Use `timeout-minutes: 5` for qui
 
 ### Cross-Run Memory Patterns
 
-gh-aw provides three built-in memory mechanisms (prefer these over `@modelcontextprotocol/server-memory` which is ephemeral per run):
+gh-aw provides two built-in memory mechanisms (prefer these over `@modelcontextprotocol/server-memory` which is ephemeral per run):
 
 | Mechanism | Persistence | Scope | Use Case |
 |-----------|------------|-------|----------|
-| `repo-memory:` | Git-backed, permanent | Cross-workflow via shared branch | Shared state across workflows (e.g., dedup indexes, quality scores) |
-| `cache-memory:` | GitHub Actions cache, ~7 days | Per-workflow or shared via key | Session state, intermediate results |
+| `cache-memory:` | GitHub Actions cache, configurable via `retention-days` (eviction-dependent) | Per-workflow or shared via key | Session state, intermediate results, cross-run context |
 | `cache:` | GitHub Actions cache | Per-workflow | Dependencies, build artifacts |
 
-**Recommended: `repo-memory:`** for cross-workflow context:
+**Recommended: `cache-memory:`** for cross-run context:
 ```yaml
 tools:
-  repo-memory:
-    branch-name: memory/my-project
-    allowed-extensions: [".md", ".json"]
-    max-file-size: 10240
-    max-file-count: 100
+  cache-memory:
+    key: news-${{ github.workflow }}-${{ inputs.article_date || 'today' }}
+    retention-days: 14
 ```
 
 **Avoid:** `@modelcontextprotocol/server-memory` — dies when the process ends, lost every run. Also skip `@modelcontextprotocol/server-sequential-thinking` — Claude has native CoT reasoning; it wastes context tokens.

.github/skills/github-agentic-workflows/SKILL.md

@@ -665,9 +665,6 @@ tools:
   github:
   cache-memory:
     id: metrics-tracking
-  repo-memory:
-    id: historical-data
-    branch: data/metrics
 ---
 
 Track metrics over time using memory:
@@ -680,8 +677,7 @@ Track metrics over time using memory:
 ```
 
 **Memory Types:**
-- **cache-memory**: 7-day retention (GitHub Actions cache)
-- **repo-memory**: Unlimited retention (Git branch)
+- **cache-memory**: Configurable retention via `retention-days` (GitHub Actions cache, eviction-dependent availability)
 
 **Use Cases:**
 - Incremental data processing
@@ -898,21 +894,6 @@ Update calculations.
 Store results back to cache-memory.
 ```
 
-#### Repo Memory (Unlimited retention)
-
-```yaml
-tools:
-  repo-memory:
-    id: long-term-metrics
-    branch: data/metrics
-```
-
-**Features:**
-- Permanent Git branch storage
-- Automatic conflict resolution
-- Full Git history
-- Accessible at `/tmp/gh-aw/repo-memory-{id}/`
-
 ### Concurrency Control
 
 Limit simultaneous runs: