Claude Skill

recall

Recall prior work from past sessions — how a bug was fixed, what was decided, where a pattern lives. Use when asked 'have we done this before' or 'how did I fix X' in Elixir/Phoenix work. ccrider MCP when available, else git + solution docs.

LLM Mart · 0 points · 0 0 0

Virus-scanned Reviewed automatically before listing.

Full trust report

Download oliver-kriska-claude-elixir-phoenix-plugins_elixir-phoenix_skills_recall-9767a82.zip · 3 KB

oliver-kriska/claude-elixir-phoenix 560 44 forks MIT Updated 2d ago

Part of oliver-kriska/claude-elixir-phoenix — 93 skills

Install

skills CLI npx skills add https://github.com/oliver-kriska/claude-elixir-phoenix/tree/main/plugins/elixir-phoenix/skills/recall

Claude Code

claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install oliver-kriska-claude-elixir-phoenix@llmmart

Git git clone https://github.com/oliver-kriska/claude-elixir-phoenix.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole oliver-kriska/claude-elixir-phoenix collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Recall — Session & History Archaeology

Search three evidence layers, cheapest first, to answer "have we solved this before?". Stop at the first layer that answers the question.

Usage

/phx:recall how did we fix the LiveView form that saved silently?
/phx:recall what did we decide about the billing context boundaries?
/phx:recall which library did we pick for rate limiting and why?

Iron Laws

Cheapest layer first — solution docs (local grep) → git history → session transcripts. Don't fetch sessions when a solution doc answers
ONE ccrider fetch = ONE subagent — get_session_messages responses are 3–15KB each. Spawn a subagent per session that writes a summary file and exits; NEVER batch multiple session fetches into one context
Cite the evidence — every claim names its source (solution doc path, commit hash, or session date + match snippet). No vague "we did this once"
Graceful degradation, stated plainly — if ccrider MCP is absent, say so once and use the fallback layers; never error out

Workflow

Layer 1: Compound Solution Docs (always)

Run Grep with keywords from the question over .claude/solutions/. Treat a hit here as the best answer — it was written for exactly this purpose. Present it and stop unless the user wants more.

Layer 2: Git Archaeology (always available)

git log --oneline --grep="{keyword}" -i -20      # commit messages
git log -S "{code-symbol}" --oneline -10          # when a symbol changed
git log --follow --oneline -10 -- {file}          # one file's history
git show {hash} --stat                            # inspect a candidate

Use -S (pickaxe) when the question names code; --grep when it names intent. Show matching commits with one-line context each.

Layer 3: Session Transcripts (ccrider MCP, gated)

Check for mcp__ccrider__* tools (load via ToolSearch if deferred).

If absent: report "ccrider MCP not connected — answered from solution docs + git history" and stop after Layers 1–2.

If present:

mcp__ccrider__search_sessions with the question's key phrases — returns ranked hits with session IDs and snippets
Present the top 3–5 hits (date, snippet) and pick the most relevant (or ask, if genuinely ambiguous)
Per selected session, spawn ONE subagent: "Fetch session via mcp__ccrider__get_session_messages, extract only what answers '', write ≤30 lines to .claude/recall/{id}.md" (Iron Law 2)
Read the summary files, synthesize the answer with citations

Step 4: Answer + Compound

Present the answer with its evidence trail. If the recalled knowledge was NOT already in .claude/solutions/, offer /phx:compound so the next recall stops at Layer 1.

Integration

"have we done this before?" → /phx:recall
   Layer 1 .claude/solutions/ ──hit──► answer + cite
   Layer 2 git log --grep/-S  ──hit──► answer + cite
   Layer 3 ccrider sessions (gated) ─► answer + cite → offer /phx:compound

References

${CLAUDE_SKILL_DIR}/references/archaeology-patterns.md — git pickaxe recipes, ccrider query patterns, subagent prompt template

Files (claude-elixir-phoenix)

references

archaeology-patterns.md 2.1 KB

# Archaeology Patterns

## Git recipes (Layer 2)

| Question shape | Recipe |
|----------------|--------|
| "When did we change/add X?" (code symbol) | `git log -S "X" --oneline` — pickaxe finds commits where the count of X changed |
| "Why does this line exist?" | `git log -L {start},{end}:{file}` or `git blame -w -C {file}` |
| "What happened to this file?" | `git log --follow --oneline -- {file}` (survives renames) |
| "What did we do about {topic}?" | `git log --grep="{topic}" -i --oneline` |
| "What changed between releases?" | `git log v2.10.0..v2.11.0 --oneline` |
| Inspect a candidate | `git show {hash} --stat`, then `git show {hash} -- {file}` for the diff |

Prefer `--oneline` + targeted `git show` over dumping full diffs into
context — same delta-only discipline as the watch-pr skill.

## ccrider query patterns (Layer 3)

- Lead with the concrete noun: `"form save silent changeset"` beats
  `"debugging problem form"`
- Search returns ranked hits with session IDs + snippets — that's usually
  enough to pick ONE session worth fetching
- `list_recent_sessions` when the user says "last week" / "yesterday"
  rather than describing content

## Subagent prompt template (Iron Law 2)

One session per subagent — `get_session_messages` responses are 3–15KB
and batching floods the parent context:

```
Fetch session {SESSION_ID} via mcp__ccrider__get_session_messages.
The question is: "{QUESTION}"
Extract ONLY content that answers it: the problem, the fix/decision,
file paths, and the outcome. Ignore everything else.
Write at most 30 lines of markdown to .claude/recall/{SESSION_ID}.md
and reply with just "done" — the file is the output.
```

Note: ccrider messages use `type` (not `role`) and content is plain
string — tool usage must be inferred from assistant text.

## When all three layers miss

Say so explicitly: "No prior work found on {topic} in solution docs, git
history, or indexed sessions." A clean miss is a valid answer — do not
pad it with speculation. Suggest `/phx:compound` after the user solves it
fresh, so the next recall hits Layer 1.

SKILL.md 3.5 KB

---
name: recall
description: Recall prior work from past sessions — how a bug was fixed, what was decided, where a pattern lives. Use when asked 'have we done this before' or 'how did I fix X' in Elixir/Phoenix work. ccrider MCP when available, else git + solution docs.
effort: medium
argument-hint: <question about past work>
---

# Recall — Session & History Archaeology

Search three evidence layers, cheapest first, to answer "have we solved
this before?". Stop at the first layer that answers the question.

## Usage

```
/phx:recall how did we fix the LiveView form that saved silently?
/phx:recall what did we decide about the billing context boundaries?
/phx:recall which library did we pick for rate limiting and why?
```

## Iron Laws

1. **Cheapest layer first** — solution docs (local grep) → git history →
   session transcripts. Don't fetch sessions when a solution doc answers
2. **ONE ccrider fetch = ONE subagent** — `get_session_messages` responses
   are 3–15KB each. Spawn a subagent per session that writes a summary
   file and exits; NEVER batch multiple session fetches into one context
3. **Cite the evidence** — every claim names its source (solution doc
   path, commit hash, or session date + match snippet). No vague "we did
   this once"
4. **Graceful degradation, stated plainly** — if ccrider MCP is absent,
   say so once and use the fallback layers; never error out

## Workflow

### Layer 1: Compound Solution Docs (always)

Run Grep with keywords from the question over `.claude/solutions/`.
Treat a hit here as the best answer — it was written for exactly this
purpose. Present it and stop unless the user wants more.

### Layer 2: Git Archaeology (always available)

```bash
git log --oneline --grep="{keyword}" -i -20      # commit messages
git log -S "{code-symbol}" --oneline -10          # when a symbol changed
git log --follow --oneline -10 -- {file}          # one file's history
git show {hash} --stat                            # inspect a candidate
```

Use `-S` (pickaxe) when the question names code; `--grep` when it names
intent. Show matching commits with one-line context each.

### Layer 3: Session Transcripts (ccrider MCP, gated)

Check for `mcp__ccrider__*` tools (load via ToolSearch if deferred).

**If absent**: report "ccrider MCP not connected — answered from solution
docs + git history" and stop after Layers 1–2.

**If present**:

1. `mcp__ccrider__search_sessions` with the question's key phrases —
   returns ranked hits with session IDs and snippets
2. Present the top 3–5 hits (date, snippet) and pick the most relevant
   (or ask, if genuinely ambiguous)
3. Per selected session, spawn ONE subagent: "Fetch session {id} via
   `mcp__ccrider__get_session_messages`, extract only what answers
   '{question}', write ≤30 lines to `.claude/recall/{id}.md`" (Iron Law 2)
4. Read the summary files, synthesize the answer with citations

### Step 4: Answer + Compound

Present the answer with its evidence trail. If the recalled knowledge was
NOT already in `.claude/solutions/`, offer `/phx:compound` so the next
recall stops at Layer 1.

## Integration

```text
"have we done this before?" → /phx:recall
   Layer 1 .claude/solutions/ ──hit──► answer + cite
   Layer 2 git log --grep/-S  ──hit──► answer + cite
   Layer 3 ccrider sessions (gated) ─► answer + cite → offer /phx:compound
```

## References

- `${CLAUDE_SKILL_DIR}/references/archaeology-patterns.md` — git pickaxe recipes, ccrider query patterns, subagent prompt template